View in EDS HTML Full Text PDF Full Text

Impact of Trial Attrition Rates on Treatment Effect Estimates in Chronic Inflammatory Diseases: A Meta-Epidemiological Study

Saved in:
Bibliographic Details
Title: Impact of Trial Attrition Rates on Treatment Effect Estimates in Chronic Inflammatory Diseases: A Meta-Epidemiological Study
Language: English
Authors: Silja H. Overgaard (ORCID 0000-0001-6184-2154), Caroline M. Moos (ORCID 0000-0003-4997-4319), John P. A. Ioannidis (ORCID 0000-0003-3118-6859), George Luta (ORCID 0000-0002-4035-7632), Johannes I. Berg, Sabrina M. Nielsen (ORCID 0000-0003-2857-2484), Vibeke Andersen (ORCID 0000-0002-0127-2863), Robin Christensen (ORCID 0000-0002-6600-0631)
Source: Research Synthesis Methods. 2024 15(4):561-575.
Availability: Wiley. Available from: John Wiley & Sons, Inc. 111 River Street, Hoboken, NJ 07030. Tel: 800-835-6770; e-mail: cs-journals@wiley.com; Web site: https://www.wiley.com/en-us
Peer Reviewed: Y
Page Count: 15
Publication Date: 2024
Document Type: Journal Articles
Reports - Research
Information Analyses
Descriptors: Epidemiology, Attrition (Research Studies), Chronic Illness, Program Effectiveness, Drug Therapy, Probability, Outcomes of Treatment
DOI: 10.1002/jrsm.1708
ISSN: 1759-2879
1759-2887
Abstract: The objective of this meta-epidemiological study was to explore the impact of attrition rates on treatment effect estimates in randomised trials of chronic inflammatory diseases (CID) treated with biological and targeted synthetic disease-modifying drugs. We sampled trials from Cochrane reviews. Attrition rates and primary endpoint results were retrieved from trial publications; Odds ratios (ORs) were calculated from the odds of withdrawing in the experimental intervention compared to the control comparison groups (i.e., differential attrition), as well as the odds of achieving a clinical response (i.e., the trial outcome). Trials were combined using random effects restricted maximum likelihood meta-regression models and associations between estimates of treatment effects and attrition rates were analysed. From 37 meta-analyses, 179 trials were included, and 163 were analysed (301 randomised comparisons; n = 62,220 patients). Overall, the odds of withdrawal were lower in the experimental compared to control groups (random effects summary OR = 0.45, 95% CI, 0.41-0.50). The corresponding overall treatment effects were large (random effects summary OR = 4.43, 95% CI 3.92-4.99) with considerable heterogeneity across interventions and clinical specialties (I2 = 85.7%). The ORs estimating treatment effect showed larger treatment benefits when the differential attrition was more prominent with more attrition in the control group (OR = 0.73, 95% CI 0.55-0.96). Higher attrition rates from the control arm are associated with larger estimated benefits of treatments with biological or targeted synthetic disease-modifying drugs in CID trials; differential attrition may affect estimates of treatment benefit in randomised trials.
Abstractor: As Provided
Entry Date: 2024
Accession Number: EJ1430334
Database: ERIC
Full text is not displayed to guests.
FullText Links:
  – Type: pdflink
    Url: https://content.ebscohost.com/cds/retrieve?content=AQICAHj0k_4E0hTGH8RJwT4gCJyBsGNe_WN95AvKlDbXJGqwxwHXQN8mEB6zh9ggNeR-Ks5KAAAA4zCB4AYJKoZIhvcNAQcGoIHSMIHPAgEAMIHJBgkqhkiG9w0BBwEwHgYJYIZIAWUDBAEuMBEEDNU646b4JWRozRRB7AIBEICBm7eHQCLyD3Xpx9NK6TycGnRCmjHB-KBuJfiClV312Uy8CjJF35T_br1bNCof_adCFvml9oNC4rTXvZMyZZjSjBtQ40Vx-U_a5m3rto2NC2u74L2jGEFlsF_G3pnQXEmi0aPla3Ynua07QmCmL8oJwuTu_aVWiNtlvKRRyouJrfEo-oZAbK34FyS3o879njh-NQh6BeHAnGMX0zcF
Text:
  Availability: 1
  Value: <anid>AN0178396405;[bdct]01jul.24;2024Jul15.06:18;v2.2.500</anid> <title id="AN0178396405-1">Impact of trial attrition rates on treatment effect estimates in chronic inflammatory diseases: A meta‐epidemiological study </title> <p>The objective of this meta‐epidemiological study was to explore the impact of attrition rates on treatment effect estimates in randomised trials of chronic inflammatory diseases (CID) treated with biological and targeted synthetic disease‐modifying drugs. We sampled trials from Cochrane reviews. Attrition rates and primary endpoint results were retrieved from trial publications; Odds ratios (ORs) were calculated from the odds of withdrawing in the experimental intervention compared to the control comparison groups (i.e., differential attrition), as well as the odds of achieving a clinical response (i.e., the trial outcome). Trials were combined using random effects restricted maximum likelihood meta‐regression models and associations between estimates of treatment effects and attrition rates were analysed. From 37 meta‐analyses, 179 trials were included, and 163 were analysed (301 randomised comparisons; n = 62,220 patients). Overall, the odds of withdrawal were lower in the experimental compared to control groups (random effects summary OR = 0.45, 95% CI, 0.41–0.50). The corresponding overall treatment effects were large (random effects summary OR = 4.43, 95% CI 3.92–4.99) with considerable heterogeneity across interventions and clinical specialties (I2 = 85.7%). The ORs estimating treatment effect showed larger treatment benefits when the differential attrition was more prominent with more attrition in the control group (OR = 0.73, 95% CI 0.55–0.96). Higher attrition rates from the control arm are associated with larger estimated benefits of treatments with biological or targeted synthetic disease‐modifying drugs in CID trials; differential attrition may affect estimates of treatment benefit in randomised trials.</p> <p>Keywords: attrition bias; biological therapy; inflammatory bowel disease; meta‐research; psoriasis; rheumatology</p> <p></p> <ulist> <item> Attrition causes missing outcome data in randomised trials carrying a risk of bias that threatens a study's internal validity.</item> <p></p> <item> Empirical estimates of the expected degree of bias in trials due to absolute and differential attrition can help interpret trial results and plan future trials.</item> <p></p> <item> Previous meta‐epidemiological studies have reported different degrees and directions of attrition and its impact on effect estimates, and previous results have been inconclusive.</item> <p></p> <item> This large sample of trials evaluating the effect of pharmacologic interventions with biological and targeted synthetic disease‐modifying drugs in chronic inflammatory diseases found, on average, evidence of strong differential attrition. Withdrawals from trials were most common among the control comparison groups.</item> <p></p> <item> Differential attrition with more attrition in the control comparators is associated with better treatment effect estimates.</item> <p></p> <item> Our findings may be universal in trials testing long‐term treatment with any pharmacologic intervention that is very effective and well tolerated, but needs confirmation in future meta‐epidemiological studies.</item> <p></p> <item> That differential attrition is widespread in CID trials stresses the importance of a thorough appraisal of the risk of attrition bias when interpreting trial results to reach reliable conclusions.</item> <p></p> <item> When planning and conducting future similar studies, researchers ought to be particularly aware of this risk of differential attrition to minimise the bias that it may cause.</item> </ulist> <p>What is Already Known What is New Potential Impact for Research Synthesis Methods Readers</p> <hd id="AN0178396405-2">INTRODUCTION</hd> <p>Randomly allocating participants to an experimental intervention or a control comparison is an integral strength of randomised controlled trials (RCTs) as it secures comparability between groups enabling the potential for causal inference. However, the credibility of RCT results is reduced if there is a risk of bias. Different methods are applied to RCTs to minimise bias and ensure comparability, such as allocation concealment and blinding of patients and assessors. Although participant withdrawal from a study is inevitable, it can jeopardise the comparability of study groups. Attrition describes the degree of withdrawal of study participants. Common reasons for attrition include not receiving the allocated treatment, loss to follow‐up, or adverse events.[<reflink idref="bib1" id="ref1">1</reflink>] Attrition introduces bias if there are systematic differences between completers and non‐completers of studies, known as selection bias.[[<reflink idref="bib1" id="ref2">1</reflink>], [<reflink idref="bib3" id="ref3">3</reflink>]] Differential attrition is when attrition rates are unequal between study arms and is of particular concern for the internal validity of an RCT. If we assume that randomisation results in comparable groups, differential attrition is likely caused by circumstances related to the randomised treatment, such as adverse events or disease deterioration.</p> <p>The intention‐to‐treat (ITT) principle is applied to preserve group comparability and limit caveats related to attrition bias. For a trial to remain causally inferential, all randomised participants must remain in the allocated group for the statistical analyses regardless of departures from randomised treatment.[<reflink idref="bib3" id="ref4">3</reflink>] The analysts need to carefully develop a strategy to handle missing data, ideally specified during the protocol stage of the study to avoid data‐driven decisions and bias. Empirical investigations of attrition bias in meta‐epidemiological studies have yielded different results depending on the disease and related interventions. Some studies have reported that attrition bias and exclusion of patients from analyses tend to result in a more beneficial effect of the intervention,[<reflink idref="bib4" id="ref5">4</reflink>] while others find inconclusive evidence[[<reflink idref="bib5" id="ref6">5</reflink>], [<reflink idref="bib7" id="ref7">7</reflink>]] or even a tendency to underestimate the treatment effect.[<reflink idref="bib8" id="ref8">8</reflink>] More evidence on the impact of attrition (absolute and differential) on bias in randomised trials is needed. This information would benefit the design of new studies and be useful for clinicians interpreting trial outcomes.</p> <p>This meta‐epidemiological study aimed to examine the impact of attrition (i.e., absolute [across groups] and differential [contrast between groups]) and reasons for attrition by focusing on RCTs investigating pharmacological interventions with biologic or targeted synthetic (small molecule) disease‐modifying drugs in chronic inflammatory diseases (CIDs). These diseases have major clinical consequences and must be managed with long‐term treatment. Therefore, attrition is an important consideration, and it is essential to understand its impact.</p> <hd id="AN0178396405-3">METHODS</hd> <p>The protocol for this study was registered at the International Prospective Register of Systematic Reviews (PROSPERO) before commencing (CRD42022361812).</p> <hd id="AN0178396405-4">Identification of meta‐analyses and randomised trials for inclusion</hd> <p>We sampled randomised trials from meta‐analyses in Cochrane reviews that synthesised evidence in accordance with the inclusion criteria (study population, experimental intervention, control comparator, outcome, and study design [PICOS]) described in Box 1. To identify these, we searched the Cochrane Database of Systematic Reviews (via PubMed) up to September 28, 2022, using the following directly applicable search terms: "<emph>Cochrane Database Syst Rev"[jour] AND (arthritis[tiab] OR spondyloarthr*[tiab] OR ankylos*[tiab] OR psoria*[tiab] OR rheumatoid[tiab] OR "inflammatory bowel" [tiab] OR Crohn*[tiab] OR "ulcerative colitis" [tiab] OR IBD[tiab])</emph>. Two reviewers (SHO and CMM) independently screened the reviews and relevant RCTs for eligibility. Disagreements were resolved by discussion or consultation with a third reviewer (RC). We used Covidence (Covidence systematic review software 2022) to manage the identified search records and document reasons for exclusion.</p> <hd id="AN0178396405-5">1 BOX Detailed description of inclusion criteria</hd> <p></p> <hd id="AN0178396405-6">Study design</hd> <p>Studies were selected with a two‐step approach. (<reflink idref="bib1" id="ref9">1</reflink>) Meta‐analyses with a benefit outcome (i.e., excluding meta‐analysis of harms and adverse events) in Cochrane reviews of randomised controlled trials were identified. (<reflink idref="bib2" id="ref10">2</reflink>) From each meta‐analysis, the component randomised trials were sampled.</p> <hd id="AN0178396405-7">Study population</hd> <p>Patients with a group of chronic inflammatory diseases that are treated with biologic and targeted synthetic disease‐modifying drugs:</p> <p></p> <ulist> <item> Gastroenterology: Crohn's disease or ulcerative colitis.</item> <p></p> <item> Rheumatology: Rheumatoid arthritis, spondyloarthritis, or psoriatic arthritis.</item> <p></p> <item> Dermatology: Psoriasis.</item> </ulist> <hd id="AN0178396405-8">Experimental intervention</hd> <p></p> <ulist> <item> Biologic disease modifying drugs (e.g., TNF inhibitors).</item> <p></p> <item> Targeted synthetic disease‐modifying drugs (small molecules).</item> </ulist> <p>Interventions were active treatment—not the tapering or withdrawal of a drug (i.e., maintenance/tapering trials were only included if the induction period was described in detail and had information about attrition etc.).</p> <hd id="AN0178396405-9">Control comparison</hd> <p></p> <ulist> <item> Placebo.</item> <p></p> <item> Non‐active comparator (e.g., methotrexate if the intervention group also received methotrexate in addition to the experimental intervention drug).</item> </ulist> <hd id="AN0178396405-10">Outcome</hd> <p></p> <ulist> <item> Benefit outcome (i.e., a measure of treatment effect; dichotomous).</item> </ulist> <hd id="AN0178396405-11">Data extraction</hd> <p>One reviewer (SHO), supported by another reviewer (CMM), extracted data using a predefined data extraction form. A random sample of 10% of the trials (i.e., 18 trials) was selected for detailed scrutiny by CMM. If information was lacking, it was sought from other related publications or databases such as clinicaltrials.gov. Data on trial characteristics, information relevant to attrition, and trial results (dichotomised treatment effects) were extracted directly from the trial publications. The decision to extract data directly from the trial reports contrasts with our original protocol intention (Appendix S2) which described extracting the effect estimates from the forest plots in the Cochrane reviews. This protocol amendment was necessary due to the inherent heterogeneity of outcomes.</p> <p>To ensure we extracted the most relevant benefit outcome from a trial, we applied the following hierarchy: (<reflink idref="bib1" id="ref11">1</reflink>) we selected the prespecified primary outcome of the study, if dichotomous, (<reflink idref="bib2" id="ref12">2</reflink>) in trials with continuous outcomes or outcomes related to safety and harms, we selected the first mentioned dichotomous benefit outcome (although with a preference for ACR20 in RA trials as this was the most commonly used primary endpoint), which is a minor deviation from the original protocol (<reflink idref="bib3" id="ref13">3</reflink>) if more time points were specified for primary outcome assessment, we selected the end of the study if it matched a time point where attrition was reported, otherwise a time point where attrition was reported. Study duration was extracted for placebo/non‐active comparator‐controlled periods. Thus, the trial duration used for this analysis may deviate from the official trial duration. We extracted risk of bias (RoB) assessments from the Cochrane review for the following domains: sequence generation, allocation concealment, blinding, and incomplete outcome data. We classified trials as overall 'low risk of bias' (coded as '0') if all domains were rated low. A trial was classified as 'unclear risk of bias' ('1') if one of the four domains was rated as unclear and none were rated as high and as 'high risk of bias' ('2') if minimum one domain was rated high. The RoB assessment was extracted from the most recent Cochrane review if trials were included in more than one meta‐analysis.</p> <hd id="AN0178396405-12">Assessment of attrition</hd> <p>For data extraction, we defined attrition as the rate of withdrawal of patients from the randomised treatment up to the time the benefit outcome was evaluated. We extracted attrition data from the total trial and each eligible treatment arm. We also extracted the number of randomised patients, the number of patients analysed, and how the studies managed missing data outcomes. Reasons for attrition were sorted into the following categories: adverse events, lack of efficacy, withdrawal of consent, protocol violation, escape to rescue therapy, lost to follow‐up, did not receive randomised treatment, administrative issues, death, pregnancy, laboratory values outside normal, physician's decision, sponsor's decision, and other/unclear. Withdrawals due to 'safety' were categorised as adverse events, and 'non‐safety' as 'others/unclear,' withdrawals due to 'patient request' were classified as 'withdrawal of consent' and 'non‐compliance' as 'protocol deviations.' Patients that entered rescue therapy or switched to open‐label treatment were considered withdrawals, although they continued the trial. If withdrawals were described with more than one reason, they were categorised as 'other/unclear.'</p> <hd id="AN0178396405-13">Statistical analysis</hd> <p>All main analyses were prespecified in the protocol (Appendix S2). Simple descriptive statistics summarised trial characteristics as frequencies, percentages, medians, and interquartile ranges. We estimated differential attrition as odds ratios (OR) by random effects meta‐analysis with restricted maximum‐likelihood estimation, coded so that OR >1 indicated more attrition in the experimental intervention. In the case of zero‐event cells, we applied a continuity correction of 0.5. Treatment effects were also estimated as OR, with OR >1 indicating a beneficial experimental intervention effect. In trials with multiple experimental intervention arms, the number of patients in the control comparator group was divided by the number of experimental intervention arms to avoid inflating the trial sample size or the number of withdrawals as we have done previously.[<reflink idref="bib9" id="ref14">9</reflink>]</p> <p>We quantified the impact of differential attrition on treatment effects by a random effects meta‐regression analysis using the log‐transformed ORs for treatment effects against OR for differential attrition. Finally, we investigated the impact of different trial characteristics on differential attrition and effect estimates in stratified meta‐analyses using restricted maximum likelihood estimation. The various characteristics were examined univariately. We report the regression coefficients for continuous variables and the average OR for the different levels of the categorical moderators with corresponding 95% confidence intervals (CI), <emph>τ</emph><sups>2</sups> and p‐values. Although it is considered good practice to accompany statistical estimates with 95% CIs, STATA does not provide a direct option for displaying these for <emph>τ</emph><sups>2</sups>. To compare trial subgroups, we report the ratio of odds ratios (ROR) using the 'interaction revisited' to the difference between two logOR estimates from the stratified meta‐analyses.[<reflink idref="bib10" id="ref15">10</reflink>] In the meta‐regression analyses, we compare trial subgroups by reporting ORs. We performed all analyses with the statistical software Stata/IC 16.1 using the 'meta' and 'metan' packages.</p> <hd id="AN0178396405-14">Patient and public involvement</hd> <p>Patients or members of the public were not involved in the planning, analyses, interpretation, or writing of the manuscript. This study addresses methodological and interpretation challenges and is not directly dependent on patients' preferences, priorities, or experiences.</p> <hd id="AN0178396405-15">RESULTS</hd> <p></p> <hd id="AN0178396405-16">Study selection and characteristics</hd> <p>We identified 435 systematic Cochrane reviews with meta‐analyses, of which 37 were eligible (Figure 1). From the 37 meta‐analyses, we identified 314 trials. After a full‐text assessment, 179 trials (331 randomised comparisons; <emph>n</emph> = 66,871 patients) were eligible, and 163 trials (301 randomised comparisons; <emph>n</emph> = 62,220 patients) were included in the in‐depth quantitative analyses as they had data on attrition available for experimental intervention and control comparator groups. Table 1 shows the reported characteristics of all the eligible trials and the trials with attrition data for individual treatment arms. Of all trials, most trials (108; 60.3%) investigated rheumatic diseases, and the most frequent experimental intervention was biologic disease‐modifying drugs (168; 93.9%), with placebo as the control comparator (170; 95%). Most trials were published after the year 2000 (170; 95%). The median proportion of women in the trials was 73% (IQR, 40.7–80.3), and the participants' mean age was 47.3 years (SD 6.3). The characteristics of the trials included in the in‐depth analyses were fairly similar to the total sample (Table 1). Of the 179 included trials, 170 (95%) reported some attrition, and 98 trials (55%) reported total attrition rates above 10% (Table 1). Attrition rates ranged from 0% to 73%, with a median of 11%. Missing data were most frequently imputed by non‐responder imputation (113 trials; 63.1%), followed by the last observation carried forward (33 trials; 18.4%). Information on attrition per arm was available for 301 comparisons: the attrition rate ranged from 0% to 68% in the experimental intervention groups (median 9%) and between 0% to 92% in the control comparator groups (median 15%). In Figure 2, the total trial attrition rate and the attrition rate per week are depicted according to CID diagnosis and the overall risk of bias judgement.</p> <p> <img src="https://imageserver.ebscohost.com/img/embimages/rdk/BDCT/01jul24/jrsm1708-fig-0001.jpg?ephost1=dGJyMNHX8kSepq84v%2bvlOLCmsE6epq5Srqa4SK6WxWXS" alt="jrsm1708-fig-0001.jpg" title="1 Flow of the trial identification and selection process. K, number of studies/trials/comparisons; SR, systematic review; RCT, randomised controlled trial." /> </p> <p></p> <p>1 TABLE Summary of trial characteristics.</p> <p> <ephtml> <table><thead valign="bottom"><tr><th align="left">Characteristics</th><th align="left">All eligible trials (<italic>k</italic> = 179)</th><th align="left">Trials with attrition data for individual arms (<italic>k</italic> = 163)</th></tr></thead><tbody valign="top"><tr><td align="left">Proportion of women, median (IQR)</td><td align="char" char="(">73<xref ref-type="fn" rid="tfn2" /> (40.7;80.3)</td><td align="char" char="(">73 (40.1;80.5)</td></tr><tr><td align="left">Age, mean (SD)</td><td align="char" char="(">47.3 (6.3)</td><td align="char" char="(">47.4 (6.2)</td></tr><tr><td align="left">Study duration (weeks), median (IQR)</td><td align="char" char="(">24 (12;30)</td><td align="char" char="(">24 (12;30)</td></tr><tr><td align="left">Clinical specialty</td></tr><tr><td align="left">Gastroenterology</td><td align="char" char="(">37 (20.7%)</td><td align="char" char="(">33 (20.3%)</td></tr><tr><td align="left">Rheumatology</td><td align="char" char="(">108 (60.3%)</td><td align="char" char="(">98 (60.1%)</td></tr><tr><td align="left">Dermatology</td><td align="char" char="(">34 (19.0%)</td><td align="char" char="(">32 (19.6%)</td></tr><tr><td align="left">Intervention</td></tr><tr><td align="left">Biologic disease‐modifying drugs</td><td align="char" char="(">168 (93.9%)</td><td align="char" char="(">154 (94.5%)</td></tr><tr><td align="left">Targeted synthetic disease‐modifying drugs</td><td align="char" char="(">11 (6.1%)</td><td align="char" char="(">9 (5.5%)</td></tr><tr><td align="left">Comparator</td></tr><tr><td align="left">Placebo</td><td align="char" char="(">170 (95.0%)</td><td align="char" char="(">157 (96.3%)</td></tr><tr><td align="left">Non‐active</td><td align="char" char="(">9 (5.0%)</td><td align="char" char="(">6 (3.7%)</td></tr><tr><td align="left">No of patients randomly assigned</td></tr><tr><td align="left"><100</td><td align="char" char="(">36 (20.1%)</td><td align="char" char="(">30 (18.4%)</td></tr><tr><td align="left">100–500</td><td align="char" char="(">92 (51.4%)</td><td align="char" char="(">86 (52.8%)</td></tr><tr><td align="left">>500</td><td align="char" char="(">51 (28.5%)</td><td align="char" char="(">47 (28.8%)</td></tr><tr><td align="left">Patients discontinuing the trials (attrition rate)</td></tr><tr><td align="left">0</td><td align="char" char="(">9 (5.0%)</td><td align="char" char="(">5 (3.1%)</td></tr><tr><td align="left"><10%</td><td align="char" char="(">72 (40.2%)</td><td align="char" char="(">67 (41.1%)</td></tr><tr><td align="left">10%–20%</td><td align="char" char="(">44 (24.6%)</td><td align="char" char="(">41 (25.2%)</td></tr><tr><td align="left">>20%</td><td align="char" char="(">54 (30.2%)</td><td align="char" char="(">50 (30.7%)</td></tr><tr><td align="left">Management of missing data</td></tr><tr><td align="left">No missing data</td><td align="char" char="(">5 (2.8%)</td><td align="char" char="(">5 (3.1%)</td></tr><tr><td align="left">Exclusion (available cases only)</td><td align="char" char="(">2 (1.1%)</td><td align="char" char="(">2 (1.2%)</td></tr><tr><td align="left">Last observation carried forward</td><td align="char" char="(">33 (18.4%)</td><td align="char" char="(">29 (17.8%)</td></tr><tr><td align="left">Non‐responder imputation</td><td align="char" char="(">113 (63.1%)</td><td align="char" char="(">107 (65.6%)</td></tr><tr><td align="left">Multiple imputation/Mixed model</td><td align="char" char="(">8 (4.5%)</td><td align="char" char="(">8 (4.9%)</td></tr><tr><td align="left">Not reported/unclear</td><td align="char" char="(">15 (8.4%)</td><td align="char" char="(">12 (7.4%)</td></tr><tr><td align="left">N.A.</td><td align="char" char="(">3 (1.7%)</td><td align="char" char="(">0 (0.0%)</td></tr><tr><td align="left">Overall RoB judgement</td></tr><tr><td align="left">Low</td><td align="char" char="(">67 (37.4%)</td><td align="char" char="(">63 (38.7%)</td></tr><tr><td align="left">High</td><td align="char" char="(">40 (22.3%)</td><td align="char" char="(">33 (20.2%)</td></tr><tr><td align="left">Unclear</td><td align="char" char="(">72 (40.2%)</td><td align="char" char="(">67 (41.1%)</td></tr><tr><td align="left">Year of publication</td></tr><tr><td align="left">≤2000</td><td align="char" char="(">9 (5.0%)</td><td align="char" char="(">8 (4.9%)</td></tr><tr><td align="left">2001–2009</td><td align="char" char="(">84 (46.9%)</td><td align="char" char="(">77 (47.2%)</td></tr><tr><td align="left">≥2010</td><td align="char" char="(">86 (48.0%)</td><td align="char" char="(">78 (47.9%)</td></tr></tbody></table> </ephtml> </p> <p>1 Abbreviations: IQR, interquartile range; K, number of trials; N.A., not applicable (information on treatment benefit was not extracted from the trial); RoB, Risk of bias; SD, standard deviation.</p> <p>2 a Missing information from two trials.</p> <p> <img src="https://imageserver.ebscohost.com/img/embimages/rdk/BDCT/01jul24/jrsm1708-fig-0002.jpg?ephost1=dGJyMNHX8kSepq84v%2bvlOLCmsE6epq5Srqa4SK6WxWXS" alt="jrsm1708-fig-0002.jpg" title="2 Descriptive box plots. Total attrition rate for the different chronic inflammatory diseases (a). Trial attrition rate per week for the different chronic inflammatory diseases (b). Total attrition rate for trials with different overall risk of bias judgements (c), and trial attrition per week for trials with different overall risk of bias judgements (d). The median and 25th and 75th percentiles are presented." /> </p> <p></p> <hd id="AN0178396405-19">Quantitative evidence synthesis</hd> <p>Figure 3 presents the two meta‐analyses of differential attrition and treatment effect for 301 and 327 randomised comparisons, respectively. Fewer patients withdrew from the trials in the experimental intervention groups than in the control comparator groups (Figure 3a), with an odds ratio for differential attrition of 0.45 (95% CI, 0.41–0.50). Thus, overall, the odds for withdrawing were approximately double in the control comparator groups compared to the experimental group. However, the heterogeneity between the trials for differential attrition was considerable (<emph>I</emph><sups>2</sups> = 63%). The meta‐analysis of treatment effect included 327 comparisons (Figure 3b). Overall, patients treated with the disease‐modifying drugs had more than four times higher odds (OR = 4.43, 95% CI 3.92–4.99) of treatment benefit (i.e., clinical remission, clinical response, or low disease activity) compared to patients treated with control comparators (Figure 3b). Although heterogeneity was high (<emph>I</emph><sups>2</sups> = 85.7%), it was explained mainly by the sizable differences in treatment effects between the clinical specialties ranging from an OR of 1.97 (95% CI 1.69–2.31) for gastroenterology trials to an OR of 24.06 (95% CI 19.37–29.89) for psoriasis trials.</p> <p> <img src="https://imageserver.ebscohost.com/img/embimages/rdk/BDCT/01jul24/jrsm1708-fig-0003.jpg?ephost1=dGJyMNHX8kSepq84v%2bvlOLCmsE6epq5Srqa4SK6WxWXS" alt="jrsm1708-fig-0003.jpg" title="3 Meta‐analyses. (a) Differential attrition comparing intervention and comparison arms (OR, 95% CI), 301 comparisons are included. OR >1 indicates more attrition in the experimental interventions compared to control comparator groups (OR = 0.45, 95% CI 0.41–0.50, I2 = 62.76%, τ2 = 0.326). (b) Overall treatment effect stratified by clinical specialty (OR, 95% CI), 327 comparisons are included. An OR >1 indicates more responders in the experimental interventions compared to control comparator groups (OR = 4.43, 95% CI 3.92 to 4.99, I2 = 85.70%, τ2 = 0.909). Comparisons are included in the figure in chronological order based on the year of publication. CI, confidence interval; OR, odds ratio." /> </p> <p></p> <p>The stratified meta‐analyses exploring the impact of different study characteristics on differential attrition (Table 2) showed that the degree of differential attrition in rheumatology trials was 49% higher than gastroenterology trials with a ratio of odds ratio (ROR) of 1.49 (0.60/0.40; ROR = 1.49, 1.21–1.84). Trials rated as 'unclear risk of bias' for the blinding domain differed significantly with a 30% lower OR estimate for differential attrition compared to trials rated as 'low risk' (OR = 0.70, 0.54 to 0.90, <emph>p</emph> = 0.006). The most common reasons for attrition among patients randomised to experimental interventions were 'adverse events' (3.3% of all patients randomised to an experimental intervention) and 'other/unclear' (3.0%) (Table 3). In contrast, the most common reasons for attrition from patients in the control comparator group were 'lack of efficacy' (7.3% of all patients randomised to the control comparator) and 'escape to rescue therapy' (7.5%).</p> <p>2 TABLE Stratified meta‐analysis: association between trial characteristics and differential attrition.</p> <p> <ephtml> <table><thead valign="bottom"><tr><th align="left">Variable</th><th align="left">Comparisons</th><th align="left">Estimate</th><th align="left">95% CI</th><th align="left"><italic>τ</italic><sup>2</sup></th><th align="left"><italic>p</italic> value</th></tr></thead><tbody valign="top"><tr><td align="left">All comparisons</td><td align="char" char=".">301</td><td align="left">OR = 0.45</td><td align="left">0.41, 0.50</td><td align="char" char=".">0.326</td><td align="char" char="."><0.001</td></tr><tr><td align="left">Clinical specialty</td><td align="left" /><td align="left" /><td align="left" /><td align="left" /><td align="char" char="."><0.001</td></tr><tr><td align="left">Rheumatology</td><td align="char" char=".">176</td><td align="left">OR = 0.40</td><td align="left">0.35, 0.46</td><td align="char" char=".">0.455</td><td align="left" /></tr><tr><td align="left">Gastroenterology</td><td align="char" char=".">59</td><td align="left">OR = 0.60</td><td align="left">0.51, 0.70</td><td align="char" char=".">0.044</td><td align="left" /></tr><tr><td align="left">Dermatology</td><td align="char" char=".">66</td><td align="left">OR = 0.51</td><td align="left">0.44, 0.60</td><td align="char" char=".">0.033</td><td align="left" /></tr><tr><td align="left">Type of intervention<xref ref-type="fn" rid="tfn4" /></td><td align="left" /><td align="left" /><td align="left" /><td align="left" /><td align="char" char=".">0.250</td></tr><tr><td align="left">Biologic disease‐modifying drugs</td><td align="char" char=".">276</td><td align="left">OR = 0.44</td><td align="left">0.40, 0.49</td><td align="char" char=".">0.351</td><td align="left" /></tr><tr><td align="left">Targeted synthetic disease‐modifying drugs</td><td align="char" char=".">25</td><td align="left">OR = 0.52</td><td align="left">0.40, 0.68</td><td align="char" char=".">0.091</td><td align="left" /></tr><tr><td align="left">Sample size</td><td align="char" char=".">301</td><td align="left">β = 1.000</td><td align="left">1.000, 1.000</td><td align="char" char=".">0.324</td><td align="char" char=".">0.223</td></tr><tr><td align="left">Year of publication</td><td align="char" char=".">301</td><td align="left">β = 1.013</td><td align="left">0.994, 1.032</td><td align="char" char=".">0.328</td><td align="char" char=".">0.188</td></tr><tr><td align="left">Trial duration, weeks</td><td align="char" char=".">301</td><td align="left">β = 0.997</td><td align="left">0.991, 1.004</td><td align="char" char=".">0.329</td><td align="char" char=".">0.419</td></tr><tr><td align="left">Method for handling missing data<xref ref-type="fn" rid="tfn4" /></td><td align="left" /><td align="left" /><td align="left" /><td align="left" /><td align="char" char=".">0.783</td></tr><tr><td align="left">No missing data</td><td align="char" char=".">9</td><td align="left">OR = 0.70</td><td align="left">0.19, 2.62</td><td align="char" char="."><0.001</td><td align="left" /></tr><tr><td align="left">Exclusion (available cases only)</td><td align="char" char=".">2</td><td align="left">OR = 0.90</td><td align="left">0.09, 8.90</td><td align="char" char="."><0.001</td><td align="left" /></tr><tr><td align="left">Non‐responder imputation</td><td align="char" char=".">206</td><td align="left">OR = 0.45</td><td align="left">0.40, 0.49</td><td align="char" char=".">0.279</td><td align="left" /></tr><tr><td align="left">Last observation carried forward</td><td align="char" char=".">50</td><td align="left">OR = 0.45</td><td align="left">0.33, 0.61</td><td align="char" char=".">0.815</td><td align="left" /></tr><tr><td align="left">Multiple imputation/Mixed model</td><td align="char" char=".">14</td><td align="left">OR = 0.59</td><td align="left">0.39, 0.89</td><td align="char" char=".">0.109</td><td align="left" /></tr><tr><td align="left">Not reported/unclear</td><td align="char" char=".">20</td><td align="left">OR = 0.45</td><td align="left">0.31, 0.67</td><td align="char" char=".">0.144</td><td align="left" /></tr><tr><td align="left">Overall RoB judgement</td><td align="left" /><td align="left" /><td align="left" /><td align="char" char=".">0.270</td><td align="left" /></tr><tr><td align="left">Low</td><td align="char" char=".">116</td><td align="left">OR = 1</td><td align="left" /><td align="left" /><td align="left" /></tr><tr><td align="left">Unclear</td><td align="char" char=".">105</td><td align="left">OR = 0.90</td><td align="left">0.70, 1.17</td><td align="left" /><td align="char" char=".">0.446</td></tr><tr><td align="left">High</td><td align="char" char=".">80</td><td align="left">OR = 1.01</td><td align="left">0.73, 1.40</td><td align="left" /><td align="char" char=".">0.932</td></tr><tr><td align="left">Sequence generation</td><td align="left" /><td align="left" /><td align="left" /><td align="char" char=".">0.270</td><td align="left" /></tr><tr><td align="left">Low</td><td align="char" char=".">178</td><td align="left">OR = 1</td><td align="left" /><td align="left" /><td align="left" /></tr><tr><td align="left">Unclear</td><td align="char" char=".">123</td><td align="left">OR = 0.94</td><td align="left">0.75, 1.17</td><td align="left" /><td align="char" char=".">0.569</td></tr><tr><td align="left">High</td><td align="char" char=".">0</td><td align="left">n.a.</td><td align="left" /><td align="left" /><td align="left" /></tr><tr><td align="left">Allocation concealment</td><td align="left" /><td align="left" /><td align="left" /><td align="char" char=".">0.270</td><td align="left" /></tr><tr><td align="left">Low</td><td align="char" char=".">180</td><td align="left">OR = 1</td><td align="left" /><td align="left" /><td align="left" /></tr><tr><td align="left">Unclear</td><td align="char" char=".">121</td><td align="left">OR = 0.94</td><td align="left">0.75, 1.18</td><td align="left" /><td align="char" char=".">0.593</td></tr><tr><td align="left">High</td><td align="char" char=".">0</td><td align="left">n.a.</td><td align="left" /><td align="left" /><td align="left" /></tr><tr><td align="left">Blinding</td><td align="left" /><td align="left" /><td align="left" /><td align="char" char=".">0.260</td><td align="left" /></tr><tr><td align="left">Low</td><td align="char" char=".">186</td><td align="left">OR = 1</td><td align="left" /><td align="left" /><td align="left" /></tr><tr><td align="left">Unclear</td><td align="char" char=".">87</td><td align="left">OR = 0.70</td><td align="left">0.54,0.90</td><td align="left" /><td align="char" char=".">0.006</td></tr><tr><td align="left">High</td><td align="char" char=".">28</td><td align="left">OR = 0.98</td><td align="left">0.67, 1.43</td><td align="left" /><td align="char" char=".">0.911</td></tr></tbody></table> </ephtml> </p> <ulist> <item>3 <emph>Note</emph>: To investigate the association between each of the trial characteristics and the treatment effects (i.e., ORs), separate REML‐based sub‐group meta‐analyses and meta‐regression analyses with random effects were performed. The corresponding 95% confidence intervals, <emph>τ</emph><sups>2</sups> and <emph>p</emph> values are reported. The slope exp(<emph>β</emph>) should be interpreted as the proportional increase (or decrease) in the treatment effect (i.e., OR) per unit increase in the trial characteristic. A slope of <emph>β</emph> = 1 indicates no association with the treatment effect, whereas for example, a slope of <emph>β</emph> = 1.013 for year of publication means that for each year, the OR increases by a factor 1.013 (ie, 1.3%—but not significant).</item> <item>4 a Convergence not achieved during <emph>τ</emph><sups>2</sups> estimation.</item> <item>3 TABLE Reasons for discontinuations in the trials.</item> </ulist> <p> <ephtml> <table><thead valign="bottom"><tr><th align="left">Reasons for attrition</th><th align="left">Intervention groups (<italic>k</italic> = 301, <italic>n</italic> = 43,009)</th><th align="left">Comparator groups (<italic>k</italic> = 163, <italic>n</italic> = 19,211)</th></tr></thead><tbody valign="top"><tr><td align="left">All reasons</td><td align="char" char="(">5386 (12.5%)</td><td align="char" char="(">4812 (25.0%)</td></tr><tr><td align="left">Adverse events</td><td align="char" char="(">1430 (3.3%)</td><td align="char" char="(">531 (2.8%)</td></tr><tr><td align="left">Lack of efficacy</td><td align="char" char="(">951 (2.2%)</td><td align="char" char="(">1407 (7.3%)</td></tr><tr><td align="left">Withdrawal of consent</td><td align="char" char="(">398 (0.9%)</td><td align="char" char="(">276 (1.4%)</td></tr><tr><td align="left">Protocol violation</td><td align="char" char="(">198 (0.5%)</td><td align="char" char="(">84 (0.4%)</td></tr><tr><td align="left">Escape to rescue therapy</td><td align="char" char="(">861 (2.0%)</td><td align="char" char="(">1433 (7.5%)</td></tr><tr><td align="left">Lost to follow‐up</td><td align="char" char="(">131 (0.3%)</td><td align="char" char="(">81 (0.4%)</td></tr><tr><td align="left">Did not receive randomised treatment</td><td align="char" char="(">74 (0.2%)</td><td align="char" char="(">34 (0.2%)</td></tr><tr><td align="left">Administrative issues</td><td align="char" char="(">10 (0.0%)</td><td align="char" char="(">7 (0.0%)</td></tr><tr><td align="left">Death</td><td align="char" char="(">26 (0.1%)</td><td align="char" char="(">9 (0.0%)</td></tr><tr><td align="left">Pregnancy</td><td align="char" char="(">9 (0.0%)</td><td align="char" char="(">0 (0.0%)</td></tr><tr><td align="left">Laboratory values outside normal</td><td align="char" char="(">4 (0.0%)</td><td align="char" char="(">2 (0.0%)</td></tr><tr><td align="left">Physician's decision</td><td align="char" char="(">19 (0.0%)</td><td align="char" char="(">6 (0.0%)</td></tr><tr><td align="left">Sponsor's decision</td><td align="char" char="(">0 (0.0%)</td><td align="char" char="(">4 (0.0%)</td></tr><tr><td align="left">Other/unclear<xref ref-type="fn" rid="tfn6" /></td><td align="char" char="(">1275 (3.0%)</td><td align="char" char="(">938 (4.9%)</td></tr></tbody></table> </ephtml> </p> <ulist> <item>5 <emph>Note</emph>: Values are numbers (percentages) across all comparisons.</item> <item>6 a The "Other/unclear" category encompasses both discontinuations reported as "Other" in the trials but also patients where the reasons for discontinuations were not clearly provided. K, number of comparisons, n, number of participants.</item> </ulist> <hd id="AN0178396405-21">The impact of attrition on treatment effects</hd> <p>In the meta‐regression analysis to explore the impact of differential attrition on treatment effects (Figure 4), we found that the two measures were significantly associated (OR = 0.73, 95% CI 0.55–0.96, <emph>p</emph> = 0.026). For an increase of one in OR differential attrition, the treatment effect decreased by 27%. In comparison with trials where most attrition occurred in the control comparator group (<emph>k</emph> = 257), the trials with most attrition in the experimental intervention group (<emph>k</emph> = 42) had 28% decreased OR of treatment effect (ROR = 0.72, 95% CI 0.50–1.04), as seen in Table S1 (Appendix S1). This table presents the results from stratified meta‐analyses of the impact of other trial characteristics on treatment effects. In addition to clinical specialty, the method for handling missing data, the risk of bias due to sequence generation, allocation concealment and blinding seemed to be significantly associated with treatment effects (Table S1). Trials judged as having 'unclear risk' of bias for sequence generation, allocation concealment and the blinding domain all had higher treatment effect estimates compared to trials judged as a 'low risk' of bias (OR = 1.18, 1.01–1.38; <emph>p</emph> = 0.033, OR = 1.26, 1.08–1.48; <emph>p</emph> = 0.004 and OR = 1.36, 1.14–1.62; <emph>p</emph> = 0.001, respectively).</p> <p> <img src="https://imageserver.ebscohost.com/img/embimages/rdk/BDCT/01jul24/jrsm1708-fig-0004.jpg?ephost1=dGJyMNHX8kSepq84v%2bvlOLCmsE6epq5Srqa4SK6WxWXS" alt="jrsm1708-fig-0004.jpg" title="4 Meta‐regression relation between log odds ratios of treatment effects and odds ratios of differential attrition. The size of every 'bubble' is proportional to the study weight (random effects). ORs for differential attrition below one indicate more attrition in control comparator groups compared to experimental interventions (OR = 0.73, 95% CI 0.55–0.96, τ2 = 0.882, p = 0.026). CI, confidence interval." /> </p> <p></p> <p>The absolute trial attrition rate and the independent attrition rates of the experimental intervention and control comparator groups were all inversely associated with estimates of treatment effects. The absolute attrition rate was associated with a significant decrease in the OR for treatment effect of 2% (OR = 0.98, 0.97–0.99, <emph>p</emph> < 0.001), 3% (OR = 0.97, 0.96– 0.98, <emph>p</emph> < 0.001) and 1% (OR = 0.99, 0.99–1.00, <emph>p</emph> = 0.002) per 1% increase in rates of absolute trial attrition, attrition in experimental intervention and attrition in control comparison, respectively (Figure S1).</p> <hd id="AN0178396405-23">DISCUSSION</hd> <p></p> <hd id="AN0178396405-24">Principal findings</hd> <p>In this meta‐epidemiological study of 179 randomised trials eligible from 37 meta‐analyses (163 trials included in quantitative synthesis), we found that more patients withdrew from the control comparator groups than experimental intervention groups resulting in a significant degree of differential attrition. Differential attrition plays an important role in a trial's internal validity as it is likely caused by circumstances related to the randomised intervention or the outcome. This study indicates that differential attrition caused by more withdrawals in the control comparator groups is associated with better treatment effect estimates. Trials with the most attrition in the experimental interventions group produce smaller treatment effect estimates.</p> <p>Attrition causes missing outcome data, which may bias the results. However, missing outcome data will not bias the results if the 'missingness' is unrelated to the outcome within each treatment group.[[<reflink idref="bib2" id="ref16">2</reflink>], [<reflink idref="bib11" id="ref17">11</reflink>]] In contrast, missing outcome data can and often will lead to bias if the 'missingness' in the outcome depends on the treatment arm AND the actual value of the outcome, especially if the effect of the experimental intervention differs from that of the control comparator.[<reflink idref="bib11" id="ref18">11</reflink>] Differential attrition between the experimental intervention and control comparator groups signals that the missingness of the outcome data is linked to the outcome, especially as we see that the reasons for missing data also differ between the groups.[<reflink idref="bib11" id="ref19">11</reflink>] In this study, 82% of the trials employed a single imputation method (non‐responder imputation or last‐observation‐carried‐forward imputation) to account for the missing data. Applying non‐responder imputation seems like a reasonable assumption for missing outcome data in the control comparator groups, as most attrition was classified as 'lack of efficacy' and 'escape to rescue therapy'. In other words, the assumption about the missing data would be that attrition rates are an inverse proxy for treatment response. However, this method fails to account for the uncertainty about the imputation yielding spurious confidence intervals and conclusions about the statistical significance.[<reflink idref="bib13" id="ref20">13</reflink>] Furthermore, "non‐responses" did not account for all the withdrawals, and other reasons dominated primarily in the experimental intervention group. If attrition depends on the outcome's true value, bias is not likely to be eliminated regardless of the imputation method applied, and the imputation method may further introduce bias. Indeed, an analysis of 235 RCTs found that different assumptions about missing outcome data from participants lost to follow‐up resulted in the loss of significance (i.e., could change the interpretation) in up to 58% of the trials.[<reflink idref="bib14" id="ref21">14</reflink>] However, if the amount of missing data (differential or not) is very low, it is unlikely that the missing outcomes will make an important difference to the estimated experimental intervention effect.[<reflink idref="bib2" id="ref22">2</reflink>] In our study, the stratified meta‐analysis of treatment effects indicated that the method of managing missing data is associated with the estimated treatment effects. Trials that applied multiple imputations or mixed methods reported the most beneficial odds of treatment effects, followed by non‐responder imputation.</p> <p>This study shows differences in the treatment effects of pharmacological interventions in various chronic inflammatory diseases: the OR for the treatment effect of biologic or targeted synthetic disease‐modifying drugs in psoriasis trials was more than 24 compared to 2 and 3 for gastroenterology and rheumatology trials, respectively. The substantial variation in effect sizes may explain the difference in differential attrition between specialties. There is a higher incentive for patients to remain in the experimental treatment arm compared to the control comparison arm of a trial when the intervention is effective. In other words, the identified difference in attrition rates indicates that patients randomised to control comparisons are, reasonably, more disappointed with the treatment and, therefore, less motivated. In addition, there are differences in research traditions between specialties that may contribute to differential attrition. For example, it is common practice in rheumatology trials to incorporate a prespecified time point for escape‐to‐rescue therapy. In the included trials, this time point was often before the primary outcome was assessed and/or before reporting withdrawals, enlarging the number of withdrawals in rheumatology trials.</p> <hd id="AN0178396405-25">Comparison with other studies</hd> <p>Differential attrition was extensive in our sample of RCTs. Generally, the presence and direction of differential attrition from other studies have been inconsistent. For example, no evidence of differential attrition was observed across a random sample of 100 RCTs published in general medical journals.[<reflink idref="bib15" id="ref23">15</reflink>] Similarly, RCTs focusing on lifestyle interventions did not report differential attrition.[<reflink idref="bib16" id="ref24">16</reflink>] Moreover, there is an inconsistency in the direction of differential attrition of studies reporting directly on differential attrition. For example, studies with anti‐depressant drug interventions[<reflink idref="bib17" id="ref25">17</reflink>] or health behaviour change interventions[<reflink idref="bib18" id="ref26">18</reflink>] observed more attrition in the experimental intervention groups than in the control comparator groups. In contrast, a study investigating the effect of intravitreal anti‐vascular endothelial growth factor injections (ophthalmology) reported higher attrition rates in the control comparison arms of the trial.[<reflink idref="bib19" id="ref27">19</reflink>] Thus, the level and direction of differential attrition appear highly dependent on the type of intervention, that is, patients will more likely withdraw from experimental interventions that require substantial effort and motivation or are associated with many side effects compared to experimental interventions with obvious treatment benefits.</p> <hd id="AN0178396405-26">Implications</hd> <p>An empirical assessment of the degree and direction of attrition has valuable and direct implications for the design of future trials, interpretation of trial results, and reviewers assessing the risk of bias. Our results show that it is reasonable to expect more attrition in control comparator groups than in the experimental interventions when designing future similar trials. This finding may be universal in trials testing long‐term treatment with any effective pharmacologic interventions (if adverse events are not common) but needs confirmation in future meta‐epidemiological studies. Researchers should seek to avoid attrition, using various approaches to minimise it.[<reflink idref="bib20" id="ref28">20</reflink>] However, it remains crucial to document reasons for attrition and plan sensitivity analyses with a range of assumptions about missing data to minimise the risk of attrition bias.[[<reflink idref="bib3" id="ref29">3</reflink>], [<reflink idref="bib11" id="ref30">11</reflink>]]</p> <p>Differential attrition, rather than absolute attrition, remains most critical to evaluate for its impact on effect estimates. Substantial differences in attrition between treatment arms should make us more aware of the potential for biased magnitudes of estimated treatment effects when evaluating the quality of evidence.[<reflink idref="bib11" id="ref31">11</reflink>] The average bias associated with defects in the conduct of randomised trials varies with the type of outcome. Therefore, systematic reviewers should routinely assess the risk of bias in the results of trials and should examine whether results change when meta‐analyses are limited to trials with a low risk of bias. Additionally, addressing and acknowledging missing data is essential for maintaining the integrity of a meta‐analysis including considering <emph>why</emph> data may be missing.[<reflink idref="bib13" id="ref32">13</reflink>] The fact that differential attrition is widespread in CID trials stresses the importance of a thorough appraisal of the risk of attrition bias in systematic reviews and meta‐analyses of such RCTs to ensure reliable interpretations. Furthermore, attrition bias may accumulate in the synthesis of results.[<reflink idref="bib21" id="ref33">21</reflink>] Most (72%) of the 179 trials in our sample of randomised trials were published in prominent journals with impact factors above 10 and, therefore, likely to be frequently cited and influential.</p> <p>In the Cochrane risk of bias tool, the domain for assessing the risk of attrition bias is called "missing outcome data".[<reflink idref="bib11" id="ref34">11</reflink>] However, even in Cochrane reviews, known for their methodological rigour, high inconsistencies in assessing attrition bias have been documented.[<reflink idref="bib22" id="ref35">22</reflink>] Hopefully, Cochrane's 2019 introduction of signalling questions in the RoB tool version 2.0, can help with a more consistent assessment of attrition bias. These questions query the amount of missing data, evidence that missing data did not bias results (e.g., sensitivity analyses), and whether the missing data (could have) depended on the true value of the missing data (e.g., evaluating the presence of differential attrition).[[<reflink idref="bib2" id="ref36">2</reflink>], [<reflink idref="bib11" id="ref37">11</reflink>]]</p> <hd id="AN0178396405-27">Strengths and limitations</hd> <p>Meta‐epidemiological studies are based on the calibre of published information and the quality of systematic reviews and RCTs. We determined withdrawals from treatment arms in all but 16 trials. Therefore, it is unlikely that bias was introduced due to inadequate reporting and misclassification.[<reflink idref="bib23" id="ref38">23</reflink>] Due to some remarkably large ORs in Figure 3b, we investigated whether our continuity correction attributed to this. However, it was only 7/327 (2.1%) of the comparisons where one of the arms had zero‐cells. Although all the zero‐cells were in the control groups, it is unlikely to explain the large ORs.</p> <p>While sampling RCTs through meta‐analyses in eligible Cochrane reviews will not identify all relevant RCTs—it will identify a large sample of CID RCTs. In this study, two reviewers selected trials from Cochrane reviews, but a degree of subjectivity in deciding on "the major benefit outcome" of the review (i.e., the meta‐analyses from which the RCTs were selected) cannot be completely ruled out.</p> <p>We acknowledge that the examination of the impact of differential attrition on treatment effects (Figure 4) poses some challenges in the interpretation, especially considering the inherent skewness of the OR for attrition. Thus, we also conducted an alternative analysis, presented in Figure S2, Appendix S1, depicting the relationship between log‐transformed OR for treatment effects and log‐transformed OR for attrition. However, despite the statistical merits of this method, we find it less straightforward to interpret. Therefore, in the main analyses, we did not log‐transform OR for attrition or other trial characteristics in Table 2.</p> <p>Meta‐epidemiological studies are observational and not immune to confounding,[<reflink idref="bib24" id="ref39">24</reflink>] and attrition bias is unlikely to be the only factor influencing the effect estimates.[<reflink idref="bib25" id="ref40">25</reflink>] Trials with high attrition rates may also have a high risk of biases in other methodological aspects influencing effect estimates. Thus, in univariable meta‐regression analyses, we tested the influence of inadequate sequence generation, allocation concealment, and blinding. These biases have been reported to be sometimes associated with larger effect estimates in other trials[[<reflink idref="bib26" id="ref41">26</reflink>]] and were a priori identified as potential confounders. In agreement with previous studies, we expected confounding by other methodological characteristics to exaggerate the estimated treatment effects.[<reflink idref="bib28" id="ref42">28</reflink>] We found that trials judged as unclear risk of bias in the sequence generation, allocation concealment and blinding domain was associated with more beneficial effect estimates compared to <emph>low</emph> risk of bias (Table S2). We have not adjusted the analyses for a combination of potential confounders, and confounding by other unknown or unmeasured factors could have affected our results. For example, although not measured, most of these trials were probably industry‐sponsored, which may also introduce bias.[[<reflink idref="bib29" id="ref43">29</reflink>]]</p> <p>We introduced heterogeneity by including three clinical specialties (six diseases) instead of restricting the eligibility criteria to one. However, this allowed a greater sample size and more power. Furthermore, these CIDs are correlated by sharing some of the same symptoms, genetics, risk factors, and treatments.[<reflink idref="bib31" id="ref44">31</reflink>] In addition, the effect and direction of attrition bias have, to our knowledge, not been examined previously in CIDs, although they are highly relevant for diseases that often require long‐term pharmacologic treatment.</p> <hd id="AN0178396405-28">CONCLUSION</hd> <p>In this large sample of CID trials evaluating the effect of pharmacologic interventions with biological and targeted synthetic disease‐modifying drugs, we found evidence of strong differential attrition caused by higher attrition in the control comparator groups. This finding is important because the empirical evidence reveals that this differential attrition—with more attrition in the control comparator groups—corresponds to larger net benefit.</p> <hd id="AN0178396405-29">AUTHOR CONTRIBUTIONS</hd> <p> <bold>Silja H. Overgaard:</bold> Conceptualization; investigation; funding acquisition; writing – original draft; methodology; validation; visualization; writing – review and editing; software; formal analysis; project administration; data curation. <bold>Caroline M. Moos:</bold> Investigation; validation; software; methodology; writing – review and editing. <bold>John P. A. Ioannidis:</bold> Writing – review and editing; methodology; supervision. <bold>George Luta:</bold> Methodology; writing – review and editing; supervision. <bold>Johannes I. Berg:</bold> Conceptualization; writing – review and editing; methodology. <bold>Sabrina M. Nielsen:</bold> Conceptualization; methodology; writing – review and editing; supervision. <bold>Vibeke Andersen:</bold> Conceptualization; funding acquisition; writing – review and editing; supervision. <bold>Robin Christensen:</bold> Conceptualization; methodology; writing – review and editing; validation; project administration; supervision.</p> <hd id="AN0178396405-30">ACKNOWLEDGEMENTS</hd> <p>We would like to thank Professor Ian White, MRC Clinical Trials Unit at UCL, London, for valuable input on the study protocol and Andreas Kristian Pedersen for statistical inputs and discussions in the study planning.</p> <hd id="AN0178396405-31">FUNDING INFORMATION</hd> <p>The Parker Institute, Bispebjerg and Frederiksberg Hospital is supported by a core grant from the Oak Foundation (OCAY‐18‐774‐OFIL). The project has furthermore received funding from the Memorial Fund of Knud and Edith Eriksen and the Region Syddanmark. The Funders had no role in planning, conducting or interpreting the study.</p> <hd id="AN0178396405-32">CONFLICT OF INTEREST STATEMENT</hd> <p>All authors have completed the ICMJE uniform disclosure form at <ulink href="http://www.icmje.org/disclosure-of-interest/">www.icmje.org/disclosure-of-interest/</ulink> and declare: no competing interests.</p> <hd id="AN0178396405-33">DATA AVAILABILITY STATEMENT</hd> <p>The data that support the findings of this study are available from the corresponding author upon reasonable request.</p> <p>GRAPH: Appendix S1: Supporting Information.</p> <p>GRAPH: Appendix S2: Supporting Information.</p> <ref id="AN0178396405-34"> <title> REFERENCES </title> <blist> <bibl id="bib1" idref="ref1" type="bt">1</bibl> <bibtext> Nunan D, Aronson J, Bankhead C. Catalogue of bias: attrition bias. BMJ Evidence‐Based Med. 2018 ; 23 (1): 21 ‐ 22. doi: 10.1136/ebmed‐2017‐110883</bibtext> </blist> <blist> <bibl id="bib2" idref="ref10" type="bt">2</bibl> <bibtext> Sterne JAC, Savovic J, Page MJ, et al. RoB 2: a revised tool for assessing risk of bias in randomised trials. BMJ. 2019 ; 366 : l4898. doi: 10.1136/bmj.l4898</bibtext> </blist> <blist> <bibl id="bib3" idref="ref3" type="bt">3</bibl> <bibtext> White IR, Horton NJ, Carpenter J, Pocock SJ. Strategy for intention to treat analysis in randomised trials with missing outcome data. BMJ. 2011 ; 342 : d40. doi: 10.1136/bmj.d40</bibtext> </blist> <blist> <bibl id="bib4" idref="ref5" type="bt">4</bibl> <bibtext> Nüesch E, Trelle S, Reichenbach S, et al. The effects of excluding patients from the analysis in randomised controlled trials: meta‐epidemiological study. BMJ. 2009 ; 339 : b3244. doi: 10.1136/bmj.b3244</bibtext> </blist> <blist> <bibl id="bib5" idref="ref6" type="bt">5</bibl> <bibtext> Hewitt CE, Kumaravel B, Dumville JC, Torgerson DJ. Assessing the impact of attrition in randomized controlled trials. J Clin Epidemiol. 2010 ; 63 (11): 1264 ‐ 1270. doi: 10.1016/j.jclinepi.2010.01.010</bibtext> </blist> <blist> <bibl id="bib6" type="bt">6</bibl> <bibtext> Dossing A, Tarp S, Furst DE, et al. Modified intention‐to‐treat analysis did not bias trial results. J Clin Epidemiol. 2016 ; 72 : 66 ‐ 74. doi: 10.1016/j.jclinepi.2015.11.003</bibtext> </blist> <blist> <bibl id="bib7" idref="ref7" type="bt">7</bibl> <bibtext> Bolvig J, Juhl CB, Boutron I, et al. Some Cochrane risk‐of‐bias items are not important in osteoarthritis trials: a meta‐epidemiological study based on Cochrane reviews. J Clin Epidemiol. 2018 ; 95 : 128 ‐ 136. doi: 10.1016/j.jclinepi.2017.11.026</bibtext> </blist> <blist> <bibl id="bib8" idref="ref8" type="bt">8</bibl> <bibtext> Armijo‐Olivo S, RdC B, Ha C, et al. Are biases related to attrition, missing data, and the use of intention to treat related to the magnitude of treatment effects in physical therapy trials? A meta‐epidemiological study. Am J Phys Med Rehabil. 2022 ; 101 (6): 520 ‐ 529. doi: 10.1097/phm.0000000000001837</bibtext> </blist> <blist> <bibl id="bib9" idref="ref14" type="bt">9</bibl> <bibtext> Tarp S, Bartels EM, Bliddal H, et al. Effect of nonsteroidal antiinflammatory drugs on the C‐reactive protein level in rheumatoid arthritis: a meta‐analysis of randomized controlled trials. Arthritis Rheum. 2012 ; 64 (11): 3511 ‐ 3521. doi: 10.1002/art.34644</bibtext> </blist> <blist> <bibtext> Altman DG, Bland JM. Interaction revisited: the difference between two estimates. BMJ. 2003 ; 326 (7382): 219. doi: 10.1136/bmj.326.7382.219</bibtext> </blist> <blist> <bibtext> Higgins JP, Savovic J, Page MJ, et al. Revised cochrane risk‐of‐bias tool for randomised trials (RoB 2). Risk of Bias Tools. Cochrane ; 2019.</bibtext> </blist> <blist> <bibtext> Higgins JP, Savovic J, Page MJ, et al. Assessing risk of bias in a randomized trial. In: Higgins JPT, Thomas J, Chandler J, et al., eds. Cohrane Handbook for Systematic Reviews of Interventions Version 6.4. Cochrane; 2023.</bibtext> </blist> <blist> <bibtext> Deeks JJ, Higgins JP, DGe A. Analysing data and undertaking meta‐analyses. In: Higgins JPT, Thomas J, Chandler J, et al., eds. Cochrane Handbook for Systematic Reviews of Interventions Version 6.4. Cochrane; 2023.</bibtext> </blist> <blist> <bibtext> Akl EA, Briel M, You JJ, et al. Potential impact on estimated treatment effects of information lost to follow‐up in randomised controlled trials (LOST‐IT): systematic review. BMJ. 2012 ; 344 : e2809. doi: 10.1136/bmj.e2809</bibtext> </blist> <blist> <bibtext> Crutzen R, Viechtbauer W, Kotz D, Spigt M. No differential attrition was found in randomized controlled trials published in general medical journals: a meta‐analysis. J Clin Epidemiol. 2013 ; 66 (9): 948 ‐ 954. doi: 10.1016/j.jclinepi.2013.03.019</bibtext> </blist> <blist> <bibtext> Bevens W, Shoushtari A, Jelinek P, et al. Systematic review and meta analysis of differential attrition between active and control arms in randomized controlled trials of lifestyle interventions in chronic disease. BMC Med Res Methodol. 2021 ; 21 (1): 122. doi: 10.1186/s12874‐021‐01313‐x</bibtext> </blist> <blist> <bibtext> Sharma T, Guski LS, Freund N, Meng DM, Gøtzsche PC. Drop‐out rates in placebo‐controlled trials of antidepressant drugs: a systematic review and meta‐analysis based on clinical study reports. Int J Risk Saf Med. 2019 ; 30 (4): 217 ‐ 232. doi: 10.3233/jrs‐195041</bibtext> </blist> <blist> <bibtext> Crutzen R, Viechtbauer W, Spigt M, Kotz D. Differential attrition in health behaviour change trials: a systematic review and meta‐analysis. Psychol Health. 2015 ; 30 (1): 122 ‐ 134. doi: 10.1080/08870446.2014.953526</bibtext> </blist> <blist> <bibtext> Rosenblatt TR, Rayess N, Al‐Moujahed A, et al. Discontinuation and loss to follow‐up rates in clinical trials of intravitreal anti‐vascular endothelial growth factor injections. Graefes Arch Clin Exp Ophthalmol. 2022 ; 260 (1): 93 ‐ 100. doi: 10.1007/s00417‐021‐05246‐5</bibtext> </blist> <blist> <bibtext> Schulz KF, Grimes DA. Sample size slippages in randomised trials: exclusions and the lost and wayward. Lancet. 2002 ; 359 (9308): 781 ‐ 785. doi: 10.1016/s0140‐6736(02)07882‐0</bibtext> </blist> <blist> <bibtext> Tierney JF, Stewart LA. Investigating patient exclusion bias in meta‐analysis. Int J Epidemiol. 2005 ; 34 (1): 79 ‐ 87. doi: 10.1093/ije/dyh300</bibtext> </blist> <blist> <bibtext> Babic A, Tokalic R, Amílcar Silva Cunha J, et al. Assessments of attrition bias in Cochrane systematic reviews are highly inconsistent and thus hindering trial comparability. BMC Med Res Methodol. 2019 ; 19 (1): 76. doi: 10.1186/s12874‐019‐0717‐9</bibtext> </blist> <blist> <bibtext> Jüni P, Egger M. Commentary: empirical evidence of attrition bias in clinical trials. Int J Epidemiol. 2005 ; 34 (1): 87 ‐ 88. doi: 10.1093/ije/dyh406</bibtext> </blist> <blist> <bibtext> Christensen R, Berthelsen DB. Controversy and debate on meta‐epidemiology. Paper 3: causal inference from meta‐epidemiology: a reasonable goal, or wishful thinking? J Clin Epidemiol. 2020 ; 123 : 131 ‐ 132. doi: 10.1016/j.jclinepi.2020.03.023</bibtext> </blist> <blist> <bibtext> Sterne JA, Juni P, Schulz KF, et al. Statistical methods for assessing the influence of study characteristics on treatment effects in 'meta‐epidemiological' research. Stat Med. 2002 ; 21 (11): 1513 ‐ 1524. doi: 10.1002/sim.1184</bibtext> </blist> <blist> <bibtext> Savovic J, Turner RM, Mawdsley D, et al. Association between risk‐of‐bias assessments and results of randomized trials in Cochrane reviews: the ROBES Meta‐epidemiologic study. Am J Epidemiol. 2018 ; 187 (5): 1113 ‐ 1122. doi: 10.1093/aje/kwx344</bibtext> </blist> <blist> <bibtext> Page MJ, Higgins JP, Clayton G, et al. Empirical evidence of study design biases in randomized trials: systematic review of Meta‐epidemiological studies. PLoS One. 2016 ; 11 (7): e0159267. doi: 10.1371/journal.pone.0159267</bibtext> </blist> <blist> <bibtext> Savović J, Jones HE, Altman DG, et al. Influence of reported study design characteristics on intervention effect estimates from randomized, controlled trials. Ann Intern Med. 2012 ; 157 (6): 429 ‐ 438. doi: 10.7326/0003‐4819‐157‐6‐201209180‐00537</bibtext> </blist> <blist> <bibtext> Lundh A, Lexchin J, Mintzes B, Schroll JB, Bero L, Cochrane Methodology Review Group. Industry sponsorship and research outcome. Cochrane Database Syst Rev. 2017 ; 2 (2): Mr000033. doi: 10.1002/14651858.MR000033.pub3</bibtext> </blist> <blist> <bibtext> Nejstgaard CH, Laursen DRT, Lundh A, et al. Commercial funding and estimated intervention effects in randomized clinical trials: systematic review of meta‐epidemiological studies. Res Synth Methods. 2022 ; 13 : 214 ‐ 228. doi: 10.1002/jrsm.1611</bibtext> </blist> <blist> <bibtext> Christensen R, Heitmann BL, Andersen KW, et al. Impact of red and processed meat and fibre intake on treatment outcomes among patients with chronic inflammatory diseases: protocol for a prospective cohort study of prognostic factors and personalised medicine. BMJ Open. 2018 ; 8 (2): e018166. doi: 10.1136/bmjopen‐2017‐018166</bibtext> </blist> </ref> <aug> <p>By Silja H. Overgaard; Caroline M. Moos; John P. A. Ioannidis; George Luta; Johannes I. Berg; Sabrina M. Nielsen; Vibeke Andersen and Robin Christensen</p> <p>Reported by Author; Author; Author; Author; Author; Author; Author; Author</p> </aug> <nolink nlid="nl1" bibid="bib10" firstref="ref15"></nolink> <nolink nlid="nl2" bibid="bib11" firstref="ref17"></nolink> <nolink nlid="nl3" bibid="bib13" firstref="ref20"></nolink> <nolink nlid="nl4" bibid="bib14" firstref="ref21"></nolink> <nolink nlid="nl5" bibid="bib15" firstref="ref23"></nolink> <nolink nlid="nl6" bibid="bib16" firstref="ref24"></nolink> <nolink nlid="nl7" bibid="bib17" firstref="ref25"></nolink> <nolink nlid="nl8" bibid="bib18" firstref="ref26"></nolink> <nolink nlid="nl9" bibid="bib19" firstref="ref27"></nolink> <nolink nlid="nl10" bibid="bib20" firstref="ref28"></nolink> <nolink nlid="nl11" bibid="bib21" firstref="ref33"></nolink> <nolink nlid="nl12" bibid="bib22" firstref="ref35"></nolink> <nolink nlid="nl13" bibid="bib23" firstref="ref38"></nolink> <nolink nlid="nl14" bibid="bib24" firstref="ref39"></nolink> <nolink nlid="nl15" bibid="bib25" firstref="ref40"></nolink> <nolink nlid="nl16" bibid="bib26" firstref="ref41"></nolink> <nolink nlid="nl17" bibid="bib28" firstref="ref42"></nolink> <nolink nlid="nl18" bibid="bib29" firstref="ref43"></nolink> <nolink nlid="nl19" bibid="bib31" firstref="ref44"></nolink>
Header DbId: eric
DbLabel: ERIC
An: EJ1430334
AccessLevel: 3
PubType: Academic Journal
PubTypeId: academicJournal
PreciseRelevancyScore: 0
IllustrationInfo
Items – Name: Title
  Label: Title
  Group: Ti
  Data: Impact of Trial Attrition Rates on Treatment Effect Estimates in Chronic Inflammatory Diseases: A Meta-Epidemiological Study
– Name: Language
  Label: Language
  Group: Lang
  Data: English
– Name: Author
  Label: Authors
  Group: Au
  Data: <searchLink fieldCode="AR" term="%22Silja+H%2E+Overgaard%22">Silja H. Overgaard</searchLink> (ORCID <externalLink term="https://orcid.org/0000-0001-6184-2154">0000-0001-6184-2154</externalLink>)<br /><searchLink fieldCode="AR" term="%22Caroline+M%2E+Moos%22">Caroline M. Moos</searchLink> (ORCID <externalLink term="https://orcid.org/0000-0003-4997-4319">0000-0003-4997-4319</externalLink>)<br /><searchLink fieldCode="AR" term="%22John+P%2E+A%2E+Ioannidis%22">John P. A. Ioannidis</searchLink> (ORCID <externalLink term="https://orcid.org/0000-0003-3118-6859">0000-0003-3118-6859</externalLink>)<br /><searchLink fieldCode="AR" term="%22George+Luta%22">George Luta</searchLink> (ORCID <externalLink term="https://orcid.org/0000-0002-4035-7632">0000-0002-4035-7632</externalLink>)<br /><searchLink fieldCode="AR" term="%22Johannes+I%2E+Berg%22">Johannes I. Berg</searchLink><br /><searchLink fieldCode="AR" term="%22Sabrina+M%2E+Nielsen%22">Sabrina M. Nielsen</searchLink> (ORCID <externalLink term="https://orcid.org/0000-0003-2857-2484">0000-0003-2857-2484</externalLink>)<br /><searchLink fieldCode="AR" term="%22Vibeke+Andersen%22">Vibeke Andersen</searchLink> (ORCID <externalLink term="https://orcid.org/0000-0002-0127-2863">0000-0002-0127-2863</externalLink>)<br /><searchLink fieldCode="AR" term="%22Robin+Christensen%22">Robin Christensen</searchLink> (ORCID <externalLink term="https://orcid.org/0000-0002-6600-0631">0000-0002-6600-0631</externalLink>)
– Name: TitleSource
  Label: Source
  Group: Src
  Data: <searchLink fieldCode="SO" term="%22Research+Synthesis+Methods%22"><i>Research Synthesis Methods</i></searchLink>. 2024 15(4):561-575.
– Name: Avail
  Label: Availability
  Group: Avail
  Data: Wiley. Available from: John Wiley & Sons, Inc. 111 River Street, Hoboken, NJ 07030. Tel: 800-835-6770; e-mail: cs-journals@wiley.com; Web site: https://www.wiley.com/en-us
– Name: PeerReviewed
  Label: Peer Reviewed
  Group: SrcInfo
  Data: Y
– Name: Pages
  Label: Page Count
  Group: Src
  Data: 15
– Name: DatePubCY
  Label: Publication Date
  Group: Date
  Data: 2024
– Name: TypeDocument
  Label: Document Type
  Group: TypDoc
  Data: Journal Articles<br />Reports - Research<br />Information Analyses
– Name: Subject
  Label: Descriptors
  Group: Su
  Data: <searchLink fieldCode="DE" term="%22Epidemiology%22">Epidemiology</searchLink><br /><searchLink fieldCode="DE" term="%22Attrition+%28Research+Studies%29%22">Attrition (Research Studies)</searchLink><br /><searchLink fieldCode="DE" term="%22Chronic+Illness%22">Chronic Illness</searchLink><br /><searchLink fieldCode="DE" term="%22Program+Effectiveness%22">Program Effectiveness</searchLink><br /><searchLink fieldCode="DE" term="%22Drug+Therapy%22">Drug Therapy</searchLink><br /><searchLink fieldCode="DE" term="%22Probability%22">Probability</searchLink><br /><searchLink fieldCode="DE" term="%22Outcomes+of+Treatment%22">Outcomes of Treatment</searchLink>
– Name: DOI
  Label: DOI
  Group: ID
  Data: 10.1002/jrsm.1708
– Name: ISSN
  Label: ISSN
  Group: ISSN
  Data: 1759-2879<br />1759-2887
– Name: Abstract
  Label: Abstract
  Group: Ab
  Data: The objective of this meta-epidemiological study was to explore the impact of attrition rates on treatment effect estimates in randomised trials of chronic inflammatory diseases (CID) treated with biological and targeted synthetic disease-modifying drugs. We sampled trials from Cochrane reviews. Attrition rates and primary endpoint results were retrieved from trial publications; Odds ratios (ORs) were calculated from the odds of withdrawing in the experimental intervention compared to the control comparison groups (i.e., differential attrition), as well as the odds of achieving a clinical response (i.e., the trial outcome). Trials were combined using random effects restricted maximum likelihood meta-regression models and associations between estimates of treatment effects and attrition rates were analysed. From 37 meta-analyses, 179 trials were included, and 163 were analysed (301 randomised comparisons; n = 62,220 patients). Overall, the odds of withdrawal were lower in the experimental compared to control groups (random effects summary OR = 0.45, 95% CI, 0.41-0.50). The corresponding overall treatment effects were large (random effects summary OR = 4.43, 95% CI 3.92-4.99) with considerable heterogeneity across interventions and clinical specialties (I2 = 85.7%). The ORs estimating treatment effect showed larger treatment benefits when the differential attrition was more prominent with more attrition in the control group (OR = 0.73, 95% CI 0.55-0.96). Higher attrition rates from the control arm are associated with larger estimated benefits of treatments with biological or targeted synthetic disease-modifying drugs in CID trials; differential attrition may affect estimates of treatment benefit in randomised trials.
– Name: AbstractInfo
  Label: Abstractor
  Group: Ab
  Data: As Provided
– Name: DateEntry
  Label: Entry Date
  Group: Date
  Data: 2024
– Name: AN
  Label: Accession Number
  Group: ID
  Data: EJ1430334
PLink https://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=eric&AN=EJ1430334
RecordInfo BibRecord:
  BibEntity:
    Identifiers:
      – Type: doi
        Value: 10.1002/jrsm.1708
    Languages:
      – Text: English
    PhysicalDescription:
      Pagination:
        PageCount: 15
        StartPage: 561
    Subjects:
      – SubjectFull: Epidemiology
        Type: general
      – SubjectFull: Attrition (Research Studies)
        Type: general
      – SubjectFull: Chronic Illness
        Type: general
      – SubjectFull: Program Effectiveness
        Type: general
      – SubjectFull: Drug Therapy
        Type: general
      – SubjectFull: Probability
        Type: general
      – SubjectFull: Outcomes of Treatment
        Type: general
    Titles:
      – TitleFull: Impact of Trial Attrition Rates on Treatment Effect Estimates in Chronic Inflammatory Diseases: A Meta-Epidemiological Study
        Type: main
  BibRelationships:
    HasContributorRelationships:
      – PersonEntity:
          Name:
            NameFull: Silja H. Overgaard
      – PersonEntity:
          Name:
            NameFull: Caroline M. Moos
      – PersonEntity:
          Name:
            NameFull: John P. A. Ioannidis
      – PersonEntity:
          Name:
            NameFull: George Luta
      – PersonEntity:
          Name:
            NameFull: Johannes I. Berg
      – PersonEntity:
          Name:
            NameFull: Sabrina M. Nielsen
      – PersonEntity:
          Name:
            NameFull: Vibeke Andersen
      – PersonEntity:
          Name:
            NameFull: Robin Christensen
    IsPartOfRelationships:
      – BibEntity:
          Dates:
            – D: 01
              M: 07
              Type: published
              Y: 2024
          Identifiers:
            – Type: issn-print
              Value: 1759-2879
            – Type: issn-electronic
              Value: 1759-2887
          Numbering:
            – Type: volume
              Value: 15
            – Type: issue
              Value: 4
          Titles:
            – TitleFull: Research Synthesis Methods
              Type: main
ResultId 1