Lost to follow-up in randomized clinical trials on long-term patient management following stroke: A cross-sectional survey

Abstract

Background:

Although long-term stroke management is critically important, poor patient adherence to follow-up appointments threatens the validity of clinical trials. This cross-sectional survey aimed to identify contributing factors and potential consequences of lost to follow-up (LTFU) in long-term stroke management trials.

Methods:

We searched Medline, Embase, Web of Science, Cochrane library, and Scopus from inception to 20 August 2024 for randomized controlled trials of multimodal post-stroke care initiated within 1 year of stroke. Data on general trial and methodological characteristics were extracted. Univariable random-effects meta-regression analyses were performed to identify LTFU predictors. Furthermore, we assessed how assumptions about LTFU affected effect estimates for significant binary primary outcomes.

Results:

Among 58 eligible reports (27,575 patients and 3349 caregivers), six trials (10.3%) did not specify patient LTFU, while 8 of 17 caregiver-inclusive trials (47.1%) omitted LTFU reporting of caregivers. The median follow-up was 12 months (interquartile range (IQR): 6–12), with LTFU rates of 9.0% (IQR: 3.2–15.4%) for patients and 14.0% (IQR: 6.8–20.7%) for caregivers. Higher LTFU odds correlated with a higher proportion of females (odds ratio (OR): 2.93, 95% confidence interval (CI): 1.30–9.29) and older age (OR: 3.05, 95% CI: 1.38–9.07). Trials involving multidisciplinary rehabilitation teams showed lower LTFU (OR: 0.05, 95% CI: 0.01–0.26). When assuming different event rates for LTFU patients, 0–14.3% of significant results were no longer significant.

Conclusion:

Overall, approximately 10% of stroke trials on long-term patient management still did not report LTFU. Identified potential risk factors may provide targets to improve the continuity of stroke management within these trial settings. Attention to patient management is critical for ensuring valid trial conclusions.

Keywords

Stroke lost to follow-up rehabilitation patient compliance risk factors randomized controlled trial

Introduction

Stroke remains a major global health burden, affecting over 100 million survivors and causing 143 million disability-adjusted life-years (DALYs) annually.¹ While randomized controlled trials (RCTs) have improved acute stroke treatment, evidence for long-term management remains limited.² The World Stroke Organization recently underscored the importance of sustained post-stroke care for preventing recurrence, improving functional outcomes and quality of life.³ This has driven a notable rise in clinical trials focused on long-term stroke management.^4,5

However, these extended-duration studies face unique methodological obstacles that threaten their validity. Lost to follow-up (LTFU) emerges as a particularly concerning issue, since complete follow-up is nearly unattainable in clinical research, where LTFU rates even exceed prespecified dropout thresholds in some trials.^6,7 Such LTFU is not merely a reduction in sample size; it fundamentally undermines randomization benefits and introduces multiple biases. As a key component of attrition, LTFU constitutes one of the principal mechanisms through which attrition bias arises in randomized trials. Patients with poorer outcomes are often more likely to discontinue participation, while those from disadvantaged backgrounds face greater barriers to continued engagement.^8,9 These patterns risk distorting treatment effect estimates and reducing statistical power.¹⁰ This challenge is especially critical in post-stroke trials, where common single-blind or open-label designs allow participants and/or assessors to know treatment assignments, inevitably introducing conscious or unconscious bias into outcome measurements.¹¹ Moreover, while family caregiver LTFU is frequently overlooked, their pivotal role in rehabilitation adherence and patient outcomes warrants equal methodological consideration.¹² To address these vulnerabilities, it is essential to thoroughly examine how specific trial design elements, particularly intervention protocols and outcome measurement strategies, influence attrition dynamics in long-term stroke studies.

Compounding these challenges, investigators often misclassify LTFU and omit strategies for handling missing data, even though recommendations for categorizing and addressing LTFU exist.^13,14 This reporting gap obscures the true impact of LTFU. Despite the Consolidated Standards of Reporting Trials (CONSORT) requiring flow diagrams and intention-to-treat (ITT) analysis, studies often fail to distinguish between administrative loss (potentially random) and treatment-related attrition (likely biased).^15,16 Heterogeneity in LTFU definitions and classification across studies further compounds these inconsistencies.¹⁷ Moreover, common missing data approaches may yield biased or falsely precise results.¹⁸ For instance, conventional methods like the last observation carried forward (LOCF) may be particularly inappropriate for stroke recovery trajectories.¹⁹ Addressing these challenges requires innovative trial designs, enhanced participant retention strategies, and more appropriate statistical methods to ensure the reliability of evidence guiding long-term stroke care worldwide.

Uncertainty remains about the prevalence and effects of LTFU in long-term stroke rehabilitation trials, hindering the development of retention strategies for trial design. This cross-sectional survey aimed to examine clinical trials focused on the long-term management of stroke patients, addressing three primary aspects, including (1) to evaluate the reporting rate of LTFU and the LTFU rates in both stroke patients and their caregivers, when available; (2) to identify factors associated with LTFU; and (3) to explore the potential impact of LTFU on trial outcomes. By investigating these areas, this study aimed to generate actionable insights for improving the design and conduct of future stroke trials. A deeper understanding of LTFU is critical not only for researchers but also for clinicians and patients, as it directly influences the quality of evidence guiding clinical practice and policy decisions.

Methods

Search strategy and selection criteria

MEDLINE, Embase, Web of Science, Cochrane library, and Scopus databases were searched with the terms “stroke” OR “apoplexy” AND “multimodal care” as titles or keywords. We conducted the systematic search on 20 August 2024, covering records from each database’s inception to this date and applied standard filters to include only English-language human RCTs. The search strategy and terms are detailed in eTable 1 in Supplement. All abstracts and articles were independently reviewed by two researchers (Z.X. and Y.L.). Discrepancies were resolved through discussion and reviewed by a third researcher (P.D.). Inclusion criteria were RCTs of long-term multifaceted care with at least 3-month follow-up in post-stroke patients. These trials focused on long-term management, including secondary prevention, functional rehabilitation, and complications management after stroke. In addition, we excluded studies that focused on specific rehabilitation approaches, early supported discharge, psychological support, or individual post-stroke complications prevention or therapy thereof. We also excluded duplicates, meta-analyses, pooled analyses, reviews, and study protocols.

Data extraction

Two independent researchers (Z.X. and Y.L.) extracted data from each article using a standardized, pilot-tested data collection form. The form was finalized after evaluation with five included studies to confirm its completeness, operability, and applicability. Any discrepancies were resolved through researcher consensus, with arbitration by a third researcher (P.D.) when necessary.

For each included trial, data were extracted for the following categories, including (1) trial characteristics (author/ trial acronym, year of publication, publication rank (Journal Citation Reports), number of study centers, sponsor country, funding source, predefined dropout rate, and sample size); (2) patient characteristics (age of patient, proportion of females); (3) caregivers; (4) intervention details (intervention performer, intervention focus, and duration of follow-up); (5) outcomes (type of outcomes, significant binary primary outcome); and (6) methodological quality indicators (allocation concealment, blinding, early stop, and study protocol). The definitions of intervention focus and outcome types are detailed in eTable 2 in Supplement.

We defined LTFU as participants who were enrolled and randomized in a clinical trial but whose primary outcome data were unavailable because they could not be contacted or assessed during follow-up.²⁰ We collected the following LTFU-related data: (1) reporting of LTFU; (2) extent of LTFU; (3) reporting methods (i.e. by treatment arm or each planned follow-up, by explicit statement or flow chart); (4) reasons for LTFU; (5) assessment of the balance of baseline characteristics (i.e. LTFU group vs retained group, LTFU between different arms); (6) analytical handling methods of LTFU; and (7) discussion of potential attrition bias. The reporting rate of LTFU was quantified as the percentage of studies that reported LTFU in all included studies. For the studies that reported LTFU, the extent of LTFU rate was quantified as the percentage of patients lost in each trial.

Statistical analysis

Trial characteristics were summarized using descriptive statistics, presented as frequencies and percentages. Due to the non-normal distribution of LTFU rates, data are presented as median with interquartile range (IQR). To compare baseline characteristics between trials that reported and did not report LTFU, categorical variables were analyzed using chi-square or Fisher’s exact tests, as appropriate. Temporal trends in the reporting of LTFU and methods for handling missing data were assessed using binary logistic regression, with results expressed as odds ratios (OR) and 95% confidence intervals (CI).

To identify factors associated with LTFU rates, we performed univariable random-effects meta-regression using the “metafor” package in R software, version 4.4.1 (R Foundation for Statistical Computing, Vienna, Austria). This analysis was restricted to trials that reported LTFU data, using the log-odds of LTFU as the outcome and trial characteristics as predictors. Trials were weighted by sample size. A continuity correction of 0.5 was added to all cells for trials with zero events to permit log-odds calculation and include all available data. The log-odds of LTFU were computed as follows

\log odds = \log (\frac{t + 0.5}{n - t + 0.5})

(1)

where t = number of LTFU patients and n = total patients in the trial. Results of regression analyses were reported with OR with 95% CI. A two-tailed p < 0.05 was considered statistically significant.

To assess the potential impact of LTFU on clinical outcomes, we applied the relatively plausible assumptions proposed by Akl et al.,²¹ using the relative event incidence in LTFU patients compared with those followed up (RI_LTFU/FU), ranging from 1 to 4. In addition, four common assumptions were considered as sensitivity analyses.²² The details of these assumptions are provided in eTable 3 in the Supplement.

Results

Study selection and characteristics

From the initial 54,796 records identified, 33,961 duplicates were removed. Subsequently, 20,835 records were screened by titles and abstracts, resulting in the exclusion of 20,748 records that did not meet the inclusion criteria. After examining 87 records in full text, a total of 58 studies remained eligible for further analysis (Figure 1).

Across the 58 studies included, a total of 27,575 patients with stroke were enrolled. General characteristics of the studies included are detailed in Table 1. The median (IQR) sample size of the RCTs was 241 (85–417) patients. Of these RCTs, the study populations had a mean age of 64.6 (SD 14.11) years (33.3% female; 67.7% male). The geographic distribution of the studies was as follows: Europe (n = 24), Asia (n = 16), Oceania (n = 9), North America (n = 8), and South America (n = 1). Regarding publication characteristics, 46 of the 58 trials (79.3%) were categorized as being published in Q1-ranked journals, and 21 (36.2%) were published in the last 5 years (2020–2024). The median duration of follow-up of trials was 12 months (IQR: 6–12; range 3–60 months). The three most common outcome assessment methods were in-person clinical visit 19 (32.8%), telephone interview 18 (31.0%), and home visit 14 (24.1%). No significant differences were found between studies that reported LTFU and those that did not, suggesting no substantial selection bias based on these factors. eTable 4 in supplement provides the list of the included studies.

Table 1.

General characteristics of 58 included trials comparing trials reporting lost to follow-up (LTFU) data (n = 52) and trials without LTFU data (n = 6).

General characteristics	All trials, No. (%)	Trials with reporting LTFU data,^a No. (%)	Trials without LTFU data,^b No. (%)	p value
Number of trials	58	52	6
Publication period				0.095
Initial-2009	6 (10.3)	4 (7.7)	2 (33.3)
2010–2014	12 (20.7)	10 (19.2)	2 (33.3)
2015–2019	19 (32.8)	17 (32.7)	2 (33.3)
2020–2024	21 (36.2)	21 (40.4)	0
Publication rank (Journal Citation Reports)				0.627
1	46 (79.3)	40 (76.9)	6 (100.0)
2	4 (6.9)	4 (7.7)	0
3	7 (12.1)	7 (13.5)	0
4	0	0	0
Not included	1 (1.7)	1 (1.9)	0
Region				0.551
Europe	24 (41.4)	23 (44.2)	1 (16.7)
Asia	16 (27.6)	14 (26.9)	2 (33.3)
Oceania	9 (15.5)	7 (13.5)	2 (33.3)
North America	8 (13.8)	6 (11.5)	2 (33.3)
South America	1 (1.7)	1 (1.9)	0
Center				0.546
Single	21 (36.2)	20 (38.5)	1 (16.7)
Multicenter	37 (63.8)	32 (61.5)	5 (83.3)
Duration of follow-up, months				0.300
3–6	27 (46.6)	25 (48.1)	2 (33.3)
7–12	23 (39.7)	19 (36.5)	4 (66.7)
>12	8 (13.8)	8 (15.4)	0
Funding				0.535
Non-industry funded	51 (87.9)	46 (88.5)	5 (83.3)
Industry funded	0	0	0
Diverse funding source^c	2 (3.4)	2 (3.8)	0
Not funded	2 (3.4)	2 (3.8)	0
Not reported	3 (5.2)	2 (3.8)	1 (16.7)
Intervention performer				0.079
Nurse	11 (19.0)	10 (19.2)	1 (16.7)
Physician/ general practitioner	4 (6.9)	2 (3.8)	2 (33.3)
Therapist	13 (22.4)	13 (25.0)	0
Multidisciplinary team	22 (37.9)	20 (38.5)	2 (33.3)
Trained researchers	8 (13.8)	7 (13.5)	1 (16.7)
Intervention focus
Risk factors	38 (65.5)	35 (67.3)	3 (50.0)	0.696
Complication prevention	16 (27.6)	14 (26.9)	2 (33.3)	1.000
Education or self-management	40 (69.0)	35 (67.3)	5 (83.3)	0.736
Type of outcome measurement				0.477
In-person clinical visit	19 (32.8)	16 (30.8)	3 (50.0)
Telephone interview	18 (31.0)	16 (30.8)	2 (33.3)
Home visit	14 (24.1)	14 (26.9)	0
Tele-rehabilitation system	3 (5.2)	3 (5.8)	0
Mailed self-report	4 (6.9)	3 (5.8)	1 (16.7)
Outcome types
Risk factor control	31 (53.4)	28 (53.8)	3 (50.0)	1.000
Cardiovascular event	13 (22.4)	11 (21.2)	2 (33.3)	0.873
Quality of life	33 (56.9)	30 (57.7)	3 (50.0)	1.000
Functional outcome	33 (56.9)	31 (59.6)	2 (33.3)	0.426
Caregiver outcome	17 (29.3)	16 (30.8)	1 (16.7)	0.806

Trials with either explicit statement about LTFU (whether occurred or not) or CONSORT flow diagram showing LTFU (whether occurred or not).

Trials not reporting LTFU occurrence were neither included in the explicit statement nor in the CONSORT flow diagram.

Studies that received funding from both industry and non-industry.

Methodological quality of trials

Regarding the methodological quality, 43 (74.1%) used adequate allocation concealment (eTable 5 in supplement). It was infeasible for most of the trials (96.6%) to blind the patients or intervention performers in the trial, while 37 (63.8%) outcome adjudicators of those were blinded to minimize additional bias. Six trials (10.3%) stopped early for benefit, and 15 (25.9%) lacked an available study protocol.

Reporting, extent and handling of LTFU

Table 2 presents the percentage of trials with details on the reporting, extent, and handling of LTFU of patients. Among the 58 trials, 52 (89.7%) included either an explicit statement on LTFU, or a CONSORT flow diagram showing LTFU, or both. Of these, 6 trials (10.3%) reported no LTFU during follow-up. In addition, among the 3 trials that did not report a CONSORT flow diagram, one was published before the implementation of the 2010 CONSORT statement. The reporting of LTFU increased marginally over time (OR: 1.18, 95% CI: 1.01–1.40). A considerable variation was found in the extent of the LTFU rate (range: 0–31.4%), with an overall median of 9.0% (IQR: 3.2–15.4%). Of these, nine trials (17.3%) demonstrated a LTFU rate exceeding 20.0%, with a predominant geographical distribution in Europe (five out of nine trials). The LTFU rates did not differ significantly between intervention (median 7.8%; IQR: 2.5–16.0%) and control arms (median 9.2%; IQR: 3.0–14.0%; p for difference = 0.750).

Table 2.

Details of reporting, extent and handling regarding lost to follow-up (LTFU) of patients in included trials.

Items	No. (%)
Among all included studies (n = 58)
Explicit statement about LTFU
Yes, stating LTFU occurred	17 (29.3)
Yes, stating LTFU did not occur	6 (10.3)
Not reporting on LTFU	35 (60.3)
CONSORT flow diagram
Yes, reporting LTFU occurred	46 (79.3)
Yes, reporting LTFU did not occur	5 (8.6)
Yes, not reporting on LTFU	4 (6.9)
No	3 (5.2)
Reporting of LTFU (overall)^a	52 (89.7)
Reporting of predefined dropout rate	38 (65.5)
Predefined dropout rate^b
⩽10% to 20%	13 (34.2)
20%	18 (47.4)
>20%	7 (18.4)
LTFU exceeds predefined dropout rate^b	6 (15.8)
Among studies reporting LTFU, whether it occurred or not (n = 52)^a
Correct and clear classification of LTFU	21 (40.4)
LTFU, median (interquartile range)	9.0% (3.2%, 15.4%)
Extent of LTFU (overall)
0%	6 (11.5)
⩽10%	23 (44.2)
>10% to 20%	14 (26.9)
>20%	9 (17.3)
Separately reported for each arm	51 (98.1)
Reported at each planned follow-up	38 (73.1)
Among studies with reporting LTFU occurred (n = 46)^c
Reporting reasons for LTFU	37 (80.4)
Withdrawal of consent, median (interquartile range)	2.8% (0%, 5.9%)
Proportion of withdrawal of consent in LTFU, median (interquartile range)	24.9% (0%, 59.7%)
Assessment of baseline characteristics between LTFU and non-LTFU	7 (15.2)
Implications of LTFU discussed	7 (15.2)
Analytical approach to handle LTFU
Complete case analysis	7 (15.2)
Intention to treat (ITT)	15 (32.6)
ITT combined last observation carried forward	9 (19.6)
ITT combined multiple imputation	6 (13.0)
ITT combined expectation–maximization algorithm	1 (2.2)
ITT with missing outcomes were assumed to failure	1 (2.2)
Statistical model	5 (10.9)
Unclear	2 (4.3)

Studies with either explicit statement about LTFU (whether occurred or not) or CONSORT flow diagram showing LTFU (whether occurred or not).

Studies with reported of predefined dropout rate (n = 38).

Studies with the number of LTFU > 0.

Among the 38 studies (65.5%) that prespecified a dropout threshold, the most frequently reported rate was 20.0% (18/38, 47.4%), with 6 trials (15.8%) exceeding this threshold. While 98.1% of trials reported LTFU by study arm and 73.1% specified LTFU at each follow-up interval, only 40.4% of trials provided a correct and clear classification of LTFU. Of the 46 trials that reported LTFU events, 37 (80.4%) specified LTFU reasons. Among these, withdrawal of consent occurred at a median rate of 1.3% (IQR: 0–5.6%), and accounted for a median of 12.9% (IQR: 0–59.3%) of all LTFU cases. Seven trials (15.2%) compared baseline characteristics between retained and LTFU patients, with four only reporting no significant differences, and three providing varying levels of detailed comparisons. Those trials showed demographic factors, such as older and female, as well as poor baseline health status, may influence LTFU rates in post-stroke trials. The potential LTFU-related bias was discussed in only 15.2% of trials.

The most frequently analytical method used for handling LTFU in the primary analysis of the trials that reported LTFU occurred was ITT principle (N = 32; 69.6%), with only 17 studies (17/32, 53.1%) explicitly describing missing data imputation techniques (Table 2). Seven trials (15.2%) employed complete case analysis, while 5 (10.9%) utilized statistical models with integrated missing data handling. However, methodological transparency was lacking in 2 trials (4.3%) that omitted analytical details. No significant temporal trend was observed in the reporting of methods for handling missing data (OR: 0.93, 95% CI: 0.81–1.06).

eTable 6 summarizes the reporting and extent of LTFU among caregivers in post-stroke rehabilitation trials. Of the 58 trials analyzed, 17 (29.3%) enrolled caregivers as study participants. Among these, nearly half (n = 8, 47.1%) documented either through an explicit statement on LTFU, or a CONSORT flow diagram, or both. Across studies that reported LTFU among caregivers, the LTFU rates ranged from 0% to 34.4%, with a median of 14.0% (IQR: 6.8–20.7%).

Association between LTFU with general characteristics of trials

Figure 2 illustrates the unweighted and sample size-weighted associations between the odds of LTFU rate with the general characteristics of the 52 trials with LTFU data. Trials with ⩾45% female patients showed significantly higher LTFU odds compared to those with ⩽35% females (OR: 2.93, 95% CI: 1.30–9.29). Patients aged ⩾70 years had greater LTFU odds than those <60 years (OR: 3.05, 95% CI: 1.38–9.07). In addition, trials with diversified funding sources exhibited markedly lower ORs for odds of LTFU versus unfunded/ unreported trials (OR: 0.04, 95% CI: 0.01–0.22). Multidisciplinary teams (OR: 0.05, 95% CI: 0.01–0.26) demonstrated significantly lower LTFU rates than general practitioners. Risk factor-focused interventions and outcome assessments were associated with reduced odds of LTFU (OR: 0.14, 95% CI: 0.05–0.37 and 0.23, 95% CI: 0.08–0.63, respectively). Compared with in-person clinical visits, mailed self-reports increased LTFU risk (OR: 3.14, 95% CI: 1.19–9.30). Trials that involved caregivers and assessed caregiver outcomes as predefined outcomes had higher odds of LTFU (OR: 1.56, 95% CI: 1.12–3.48) than those without caregiver outcome assessment.

Potential impact of LTFU

Among the 52 trials reporting LTFU, 7 (13.5%) included significant binary primary outcomes suitable for evaluating the potential impact of LTFU. These outcomes were categorized into five related to risk factor control (e.g. systolic blood pressure control rate and proportion of healthy behaviors) and two related to cardiovascular events (e.g. incidence of major cardiovascular events). Detailed primary outcomes of these 7 trials can be found in eTable 7 of the supplementary materials.

Under the RI_LTFU/FU method, the proportion of eligible trials that lost significance varied according to the assumed relative event incidence between intervention and control arms, ranging from 0% to 14.3%. Among the four common assumptions evaluated in sensitivity analyses across the seven trials (eTable 8), the percentage losing statistical significance varied: 0% (0/7, best-case scenario, with no change in positive results), 14.3% (1/7, no patients LTFU had the event, with one trial showing a change), 14.3% (1/7, all patients LTFU had the event, with one trial showing a change), and 42.9% (3/7, worst-case scenario, with three trials showing a change).

Discussion

Our study provided a comprehensive evaluation of LTFU in long-term stroke rehabilitation trials, revealing critical gaps in reporting practices, methodological transparency, and potential biases that threaten the validity of trial outcomes. We found that nearly 1 in 10 trials of long-term stroke management did not report whether LTFU had occurred, though such reporting has marginally improved over time. Importantly, up to a third of the trials did not report how to handle missing data of the outcome variables for the LTFU patients, and only 15.2% considered the bias LTFU may introduce. In addition, among the trials with binary primary outcomes related to risk factor control or cardiovascular events, one of seven trials might no longer be significant if reasonable assumptions were made about the binary primary outcome in patients with LTFU. The results demand urgent attention from the stroke research community, as they directly impact the reliability of evidence guiding clinical practice for millions of stroke survivors worldwide.

The observed median LTFU rate of 9.0% (IQR: 3.2–15.4%) represents a substantial threat to trial validity, with 17.3% of studies exceeding 20%, a level generally considered critical in RCTs.²³ Reported 0% LTFU rates may raise methodological concerns, as some attrition is expected in long-term clinical trials. Such reports could suggest overly restrictive participant selection, limited follow-up duration, or potential underreporting of missed visits, rather than perfect retention. Notably, regional variation in LTFU may stem from differences in study design, participant characteristics, or healthcare infrastructure, though limited reporting of sociodemographic factors precludes firm conclusions. Beyond LTFU rates, methodological shortcomings are more concerning. The LTFU term was often misused, with some trials conflating it with other dropout categories, including mortality and health-related withdrawals. As a key component of attrition, LTFU specifically refers to participants who become unreachable after randomization, leading to missing outcome data and potential bias when loss differs by treatment group. Only 40.4% of trials correctly classified and reported LTFU, and 20.0% failed to adequately report reasons, even though those were published in high-quality journals. This methodological concern has been previously identified in similar analyses.²⁴

Reporting of baseline demographics for LTFU patients was notably limited, with merely 15.2% of trials comparing baseline characteristics between retained and LTFU patients, and fewer than half of these providing detailed data. Among the three trials lacking a CONSORT diagram, two predated the 2010 CONSORT statement, reflecting a missed opportunity to assess potential attrition bias despite its emphasis in CONSORT guidelines. Equally concerning, only 15.2% of trials discussed the potential bias introduced by LTFU, limiting readers’ ability to judge whether treatment effects were affected by selective loss. Differential LTFU between groups may introduce bias, especially if related to prognostic factors such as disease severity, which may predispose participants to withdrawal or death.²⁵ These omissions compromise the completeness of trial results and highlight the need for authors to report LTFU details, preferably in supplementary materials, and to discuss possible baseline imbalances between groups.

Our analysis identifies what is potentially the most overlooked finding in this study, which is the remarkably poor documentation of caregiver LTFU in stroke rehabilitation trials. Nearly half (47.1%) of trials involving caregivers failed to properly document LTFU, while those reporting data showed higher median LTFU (14.0%) than patient rates. This identifies a critical methodological blind spot in stroke rehabilitation research. Such neglect is particularly concerning given the well-established role of caregivers in rehabilitation adherence and long-term outcomes. The observed maximum caregiver LTFU rate of 34.4% suggests that many trials may be substantially overestimating intervention effects by failing to account for this vulnerable population’s dropout patterns. These results highlight the urgent need for dedicated CONSORT extensions to address caregiver LTFU in rehabilitation trials.

The identified predictors of LTFU provide preliminary insights that may help guide future trial design. Trials with higher female enrollment and older participants appeared more likely to experience LTFU, consistent with prior studies linking these factors to psychosocial burden and reduced follow-up adherence.^26,27 However, reducing the participation of female or older adults is not a solution, as it would compromise equity; instead, tailored strategies to better support these groups should be strengthened. Simple measures such as home visits, telephone follow-up, and tele-rehabilitation may help mitigate LTFU among older or mobility-limited patients, while mailed self-reports were associated with higher attrition.²⁸ Exploratory analyses also suggested that multidisciplinary interventions and risk factor-focused approaches were associated with lower LTFU, whereas trials including caregiver outcomes tended to have higher LTFU, possibly reflecting added study complexity.²⁹

Our results reveal a concerning stagnation in missing data methodology, with no improvement over time in reporting handling methods. While 69.6% of trials reported using ITT analysis which is recommended as the most cautious approach,³⁰ only 40% described their imputation methods. Among those that did, many relied on simplistic approaches such as LOCF, which are inappropriate for stroke recovery. The optimal approach for handling missing data remains controversial, as all methods rely on untestable assumptions.³¹ Therefore, preventing missing data with careful design and management remains essential.³² The 15.2% of trials using complete case analysis are particularly concerning, as this method assumes data are missing completely at random, an untenable assumption in most rehabilitation trials.³³ Advanced methods such as multiple imputation or mixed-effects models were seldom applied, reflecting current methodological preferences and suggesting room for further refinement in future research.

The assumptions used in this study illustrate the potential impact of LTFU on trial outcomes. Under the worst-case scenario, 42.9% of trials with initially significant results lost significance, whereas 14.3% changed under more plausible assumptions about differential event rates. These findings suggest that trial conclusions based on statistically significant binary outcomes may be sensitive to assumptions about LTFU. However, all assumptions were hypothetical and cannot be empirically verified, as the actual outcomes of patients LTFU are unknown.³⁴ Therefore, these analyses should be considered exploratory in nature, and the findings interpreted with appropriate caution within the methodological limitations of such approaches.

Strengths and limitations

This study gives a comprehensive cross-sectional survey of the LTFU-related issues in the design and statistical analysis of RCTs in long-term stroke rehabilitation, providing valuable insights into this persistent methodological challenge. We utilized rigorous survey methods for study screening and included all eligible studies to ensure a comprehensive collection of RCTs in this field. To the best of our knowledge, this is the first cross-sectional survey to exclusively assess the reporting, extent, and methodological approaches to handling LTFU in long-term stroke management RCTs, and quantify the potential impact of LTFU on binary primary outcome. The findings identify specific patient groups at higher risk of LTFU while revealing effective intervention characteristics associated with better retention. Meanwhile, we summarized the adherence of caregivers which is a critical yet frequently overlooked dimension in rehabilitation trials LTFU research.

Nevertheless, our study had several limitations. First, potential selective reporting bias may exist as unpublished LTFU analyses were unavailable. Second, even among published results, inconsistencies in reporting and classification standards may have obscured the true extent and patterns of LTFU. Third, our analysis of LTFU impact was restricted to trials reporting statistically significant binary primary outcomes, as these studies are more likely to influence clinical practice. In addition, our results do not apply to continuous data, as analyzing continuous data poses methodological challenges in missing data imputation and longitudinal interpretation. Fourth, we did not investigate the impact of LTFU on binary secondary outcomes, as primary outcomes determine trial power and clinical decision-making and secondary outcomes only provide complementary evidence. Fifth, associations between trial characteristics and LTFU rates were examined using univariable rather than multivariable meta-regression, as the limited number of eligible trials made multivariable modeling unstable and unreliable. Nonetheless, residual confounding may remain, and these findings should be interpreted with caution. Finally, although we employed varying RI_LTFU/FU ratios to explore the potential impact of LTFU on intervention effect estimates and used four common assumptions for sensitivity analyses, all these approaches are based on unverifiable assumptions. As previously discussed in the literature, such assumptions may not accurately represent the actual outcomes of patients LTFU and could lead to either overestimation or underestimation of the intervention effect. Consequently, the results should be interpreted with caution, acknowledging the inherent uncertainty of these analytical strategies in the absence of individual-level outcome data.

Conclusion

This study provides a comprehensive evaluation to date of LTFU in long-term stroke rehabilitation trials. The results reveal widespread methodological shortcomings in how LTFU is measured, reported, and analyzed which systematically compromise the reliability of stroke rehabilitation trial evidence. Importantly, our findings also provide clear, actionable pathways for improvement. By implementing rigorous LTFU monitoring, adopting advanced statistical methods, and developing targeted retention strategies, the stroke research community can enhance the reliability of rehabilitation trials. This is not merely a methodological concern, but also an ethical imperative to ensure that the millions of stroke survivors worldwide receive care based on robust, trustworthy evidence.

Supplemental Material

sj-docx-1-wso-10.1177_17474930251394853 – Supplemental material for Lost to follow-up in randomized clinical trials on long-term patient management following stroke: A cross-sectional survey

Supplemental material, sj-docx-1-wso-10.1177_17474930251394853 for Lost to follow-up in randomized clinical trials on long-term patient management following stroke: A cross-sectional survey by Peipei Du, Mingzhen Qin, Yan Liu, Xu Pang, Sijin Wang, Yixuan Li, Jierong Gao, Ziwen Xu and Chi Zhang in International Journal of Stroke

Footnotes

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Beijing Natural Science Foundation (grant no. 7252237) and the Leading Talents Project of Dongzhimen Hospital (grant no. DZMG-LJRC0008).

ORCID iDs

Peipei Du

Mingzhen Qin

Yan Liu

Xu Pang

Sijin Wang

Yixuan Li

Jierong Gao

Ziwen Xu

Chi Zhang

Supplemental material

Supplemental material for this article is available online.

References

GBD; 2019 Stroke Collaborators. Global, regional, and national burden of stroke and its risk factors, 1990-2019: a systematic analysis for the Global Burden of Disease Study 2019. Lancet Neurol 2021; 20: 795–820.

Liu

, et al. Stroke in China: advances and challenges in epidemiology, prevention, and management. Lancet Neurol 2019; 18: 394–405.

World Stroke Day. World Health Organization, https://www.who.int/southeastasia/news/detail/29-10-2020-world-stroke-day-ms (2020, accessed 8 April 2021)

Boehme

Toell

Lang

Knoflach

Kiechl

Longer term patient management following stroke: a systematic review. Int J Stroke 2021; 16: 917–926.

Sakakibara

Lear

Barr

, et al. Telehealth coaching to improve self-management for secondary prevention after stroke: a randomized controlled trial of Stroke Coach. Int J Stroke 2022; 17: 455–464.

Fewtrell

Kennedy

Singhal

, et al. How much loss to follow-up is acceptable in long-term randomised trials and prospective studies. Arch Dis Child 2008; 93: 458–461.

Day

Bierhals

CCBK

Mocellin

, et al. Nursing Home Care Intervention Post Stroke (SHARE) 1 year effect on the burden of family caregivers for older adults in Brazil: A randomized controlled trial. Health Soc Care Community 2021; 29: 56–65.

Schulz

Grimes

DA.

Sample size slippages in randomised trials: exclusions and the lost and wayward. Lancet 2002; 359: 781–785.

Valgimigli

Garcia-Garcia

Vrijens

, et al. Standardized classification and framework for reporting, interpreting, and analysing medication non-adherence in cardiovascular clinical trials: a consensus report from the Non-adherence Academic Research Consortium (NARC). Eur Heart J 2019; 40: 2070–2085.

10.

Nüesch

Trelle

Reichenbach

, et al. The effects of excluding patients from the analysis in randomised controlled trials: meta-epidemiological study. BMJ 2009; 339: b3244.

11.

Joseph

Sim

Ogollah

Lewis

A systematic review finds variable use of the intention-to-treat principle in musculoskeletal randomized controlled trials with missing data. J Clin Epidemiol 2015; 68: 15–24.

12.

Bakas

McCarthy

Miller

EL.

Systematic review of the evidence for stroke family caregiver and dyad interventions. Stroke 2022; 53: 2093–2102.

13.

Cook

Zea

Missing data and sensitivity analysis for binary data with implications for sample size and power of randomized clinical trials. Stat Med 2020; 39: 192–204.

14.

Schulz

Chalmers

Hayes

, et al. Empirical evidence of bias. Dimensions of methodological quality associated with estimates of treatment effects in controlled trials. JAMA 1995; 273: 408–412.

15.

Fong

LCW

Ford

Da Costa

, et al. Bias and loss to follow-up in cardiovascular randomized trials: a systematic review. J Am Heart Assoc 2020; 9: e015361.

16.

Butcher

Monsour

Mew

, et al. Guidelines for reporting outcomes in trial reports: the CONSORT-outcomes 2022 extension. JAMA 2022; 328: 2252–2264.

17.

Nunan

Aronson

Bankhead

Catalogue of bias: attrition bias. BMJ Evid Based Med 2018; 23: 21–22.

18.

Newgard

Lewis

RJ.

Missing data: how to best account for what is not known. JAMA 2015; 314: 940–941.

19.

Fleming

TR.

Addressing missing data in clinical trials. Ann Intern Med 2011; 154: 113–117.

20.

Fergusson

Aaron

Guyatt

, et al. Post-randomisation exclusions: the intention to treat principle and excluding patients from analysis. BMJ 2002; 325: 652–654.

21.

Akl

Briel

You

, 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.

22.

White

Horton

Carpenter

Pocock

SJ.

Strategy for intention to treat analysis in randomised trials with missing outcome data. BMJ 2011; 342: d40.

23.

Serdar

Cihan

Yücel

, et al. Sample size, power and effect size revisited: simplified and practical approaches in pre-clinical, clinical and laboratory studies. Biochem Med (Zagreb) 2021; 31: 010502.

24.

Campbell

Willox

Jhund

, et al. Reporting of lost to follow-up and treatment discontinuation in pharmacotherapy and device trials in chronic heart failure: a systematic review. Circ Heart Fail 2016; 9.

25.

Dumville

Torgerson

Hewitt

CE.

Reporting attrition in randomised controlled trials. BMJ 2006; 332: 969–971.

26.

Yoon

Bushnell

CD.

Stroke in women: a review focused on epidemiology, risk factors, and outcomes. J Stroke 2023; 25: 2–15.

27.

Vega

Benito-León

Bermejo-Pareja

, et al. Several factors influenced attrition in a population-based elderly cohort: neurological disorders in Central Spain Study. J Clin Epidemiol 2010; 63: 215–222.

28.

Chen

Sun

Zhang

, et al. Effects of home-based telerehabilitation in patients with stroke: a randomized controlled trial. Neurology 2020; 95: e2318–e2330.

29.

Grant

Hunt

Steadman

Common caregiver issues and nursing interventions after a stroke. Stroke 2014; 45: e151–3.

30.

Newell

DJ.

Intention-to-treat analysis: implications for quantitative and qualitative research. Int J Epidemiol 1992; 21: 837–841.

31.

Sidi

Harel

The treatment of incomplete data: reporting, analysis, reproducibility, and replicability. Soc Sci Med 2018; 209: 169–173.

32.

Sterne

White

Carlin

, et al. Multiple imputation for missing data in epidemiological and clinical research: potential and pitfalls. BMJ 2009; 338: b2393.

33.

Knol

Janssen

Donders

, et al. Unpredictable bias when using the missing indicator method or complete case analysis for missing confounder values: an empirical example. J Clin Epidemiol 2010; 63: 728–736.

34.

Unnebrink

Windeler

Intention-to-treat: methods for dealing with missing values in clinical trials of progressively deteriorating diseases. Stat Med 2001; 20: 3931–3946.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.06 MB