Over the past few decades, palliative oncology has evolved into a discipline that places an increased emphasis on evidence-based practice. This is supported by a growing number of clinical trials.1, 2 However, there remain many challenges to conducting palliative cancer care research, including limited research funding, difficulty in recruiting and retaining patients, few trained personnel, and other methodological issues.3-6
Because of the frail population in palliative oncology clinical studies, patient dropout is an important consideration. Indeed, attrition is a common concern among palliative care clinical trials7, 8 and contributes to selection bias, an underpowered study, and premature trial closure. However, to the best of our knowledge, there are only few studies published to date documenting the pattern of and reasons for dropout in palliative care clinical trials.9 A better understanding of the rate of dropout in palliative oncology clinical studies could help investigators to plan the appropriate sample size for grant budget and analysis purposes. Furthermore, identification of the reasons and predictive factors contributing to attrition may allow us to better design studies to improve retention, thus maximizing the number of patients with the outcome of interest assessed. In the current study, we determined the rate, reasons, and factors associated with attrition both before reaching the primary endpoint and at the end of the study.
MATERIALS AND METHODS
All 18 prospective interventional clinical trials involving patients with advanced cancer treated in the Department of Palliative Care and Rehabilitation Medicine at The University of Texas MD Anderson Cancer Center in Houston between 1999 and 2011 were included. All studies were closed to patient entry. Because this was a secondary data analysis, patients were not recontacted or reconsented. The Institutional Review Board at The University of Texas MD Anderson Cancer Center approved the study and waived the requirement for informed consent to be obtained.
Data Collection for Clinical Trial Characteristics
We retrieved information regarding various clinical trial characteristics from study protocols and published articles if available. These included the study objective (ie, symptom of interest), timing of primary outcome assessment, study design (ie, randomization, blinding, active intervention, and control intervention), study duration, study setting (inpatient or outpatient), number of study sites, funding, planned sample size, number of patients enrolled, number of dropouts, and the final sample size. Inpatient studies were those that involved hospitalized patients, whereas outpatient studies included those treated at our ambulatory clinic or during home hospice. To maximize generalizability of the study findings, clinical trials by our group and many others generally include all histologies rather than specific cancer types. Dropouts because of symptom progression or death were not considered to be protocol failures in our clinical trials.
Data Collection for Patient Characteristics
Information was collected regarding various patient characteristics at the time of clinical trial enrollment, including age, sex, race, marital status, educational level, cancer diagnosis, Eastern Cooperative Oncology Group (ECOG) performance status, and the Edmonton Symptom Assessment Scale (ESAS). ESAS is a validated scale designed to assess 10 common symptoms (ie, pain, fatigue, nausea, depression, anxiety, drowsiness, shortness of breath, appetite, feelings of well-being, and sleep). Patients were asked to rate the severity of their symptoms over the previous 24 hours using a numerical rating scale of 0 to 10, with 0 indicating that the symptom was absent and 10 indicating the worst possible symptom.10-12 Although a vast majority of studies incorporated ESAS and some also included ECOG performance status, they were not uniformly included in all studies.
All patients were classified based on whether they completed the primary outcome measure and whether they completed the entire study, and the reasons for dropout when available.
The baseline patient demographics and study characteristics were summarized using descriptive statistics, including medians, means, standard deviations, ranges, interquartile ranges (IQRs), and frequencies.
We calculated the 2 study-level outcomes, attrition rate before the primary endpoint and before the end of the study, by dividing the number of patients who dropped out before the specific time point by the total number of patients enrolled in each clinical trial. Study level predictors for the 2 outcomes were then analyzed using the Spearman correlation test for continuous variables and the Kruskal-Wallis test for categorical variables.
We determined patient-level predictors of dropout before the primary endpoint using the single-predictor random effects generalized linear mixed model (GLMM), with dropout as the dichotomous outcome, predictors as fixed effects, and study as a random effect to account for differences among clinical trials. Because several patient characteristics were not collected across all studies (eg, ECOG performance status), we applied multiple imputation techniques to the patient-level data set to produce estimates for the missing data. Overall, 19% (range, 0%-59%) of the data were missing, requiring imputation. Variables that were significant at the .10 level in single-predictor GLMM were then included in the multiple-predictor GLMM. P values of < .05 were considered to be statistically significant. A similar analysis was conducted for attrition before the end of the study.
We used IVEware2 (build 2012.02; University of Michigan, Ann Arbor, Mich) to perform the imputation using linear regression techniques. For all analyses other than imputations, we used SAS/STAT statistical software (version 9.2; SAS Inc, Cary, NC).
Clinical Trial Characteristics
Table 1 summarizes the 18 interventional clinical trials.13-20 Of these 18 trials, 15 (83%) were randomized controlled trials, and 10 of 18 (56%) included a placebo as the control. Cancer-related fatigue was the most common symptom investigated. Only 2 of the studies enrolled the planned number of subjects without dropouts.
Table 1. Clinical Trial Characteristics (n=18)
|ID00-030 Nebulized/ subcutaneous morphine15||Dyspnea||Yes||No||Yes||No||1 h||2||No||12|
|2006-0591 Interactive voice response||Symptoms||Yes||No||No||Yes||15||15||No||33|
|2006-0641 Methadone as co-opioid||Pain||Yes||No||No||Yes||15||15||No||5|
|2006-0739 Multimodal interventions||Cachexia||No||No||No||Yes||29||29||No||8|
The baseline characteristics of the 1214 cancer patients at the time of enrollment are shown in Table 2. Gastrointestinal, respiratory, and breast were the most common cancers. Approximately one-third of patients had an ECOG performance status of ≥ 3. The median symptom burden by ESAS was relatively high, particularly with regard to fatigue, anorexia, and sleep.
Table 2. Patient Characteristics (n=1214)
|Average age (range), y||60 (23–93)|
|Female sex||685 (56)|
| White||691 (57)|
| Black||153 (13)|
| Hispanic||129 (11)|
| Asian||23 (2)|
| Not available||218 (18)|
| Yes||619 (51)|
| No||312 (26)|
| Not available||283 (23)|
| High school or less||217 (18)|
| College||237 (20)|
| Advanced degree||39 (3)|
| Not available||721 (59)|
|ECOG performance status|| |
| 0||20 (2)|
| 1||184 (15)|
| 2||251 (21)|
| 3||171 (14)|
| 4||57 (5)|
| Not available||531 (44)|
| Breast||210 (17)|
| Gastrointestinal||282 (23)|
| Genitourinary||112 (9)|
| Gynecological||93 (8)|
| Head and neck||63 (5)|
| Hematological||58 (5)|
| Other||144 (12)|
| Respiratory||222 (18)|
| Not available||30 (3)|
|Median Edmonton Symptom Assessment Scale (IQR)|| |
| Pain||4 (2–7)|
| Fatigue||7 (5–8)|
| Nausea||1 (0–4)|
| Depression||2 (0–5)|
| Anxiety||2 (0–5)|
| Drowsiness||4 (2–6)|
| Appetite||5 (3–8)|
| Well-being||5 (3–7)|
| Dyspnea||2 (0–5)|
| Sleep||5 (2–7)|
| Prior to the primary endpoint||311 (26)|
| Prior to the end of study||535 (44)|
Rates and Reasons for Attrition
At the clinical trial level, the median attrition rate was 28% (IQR, 16%-38%) for the primary endpoint and 44% (IQR, 28%-58%) for the end-of-study endpoint.
At the patient level, the attrition rate was 26% (95% confidence interval [95% CI], 23%-28%) for the primary endpoint and 44% (95% CI, 41%-47%) for the end-of-study endpoint.
The main reasons for attrition were patient withdrawal and clinical deterioration, which accounted for 48% to 52% and 23% to 35%, respectively, of the dropouts (Table 3). Specifically, a high symptom burden, which may or may not be related to the clinical trial intervention, was the most common reason for patient withdrawal.
Table 3. Reasons for Attrition
|Lost to follow-up||14 (5)||22 (4)|
|Deterioration|| || |
| Death||19 (6)||45 (8)|
| Disease progression||3 (1)||5 (1)|
| Altered mental status||12 (4)||17 (3)|
| Unable to take medication orally||5 (2)||5 (1)|
| Transfer to hospice||5 (2)||12 (2)|
| Hospital admission||30 (10)||43 (8)|
|Patient withdrawal|| || |
| Patient decision||47 (15)||93 (17)|
| Symptom burden||65 (21)||87 (16)|
| Family/caregiver decision||10 (3)||18 (3)|
| Other clinical trial/therapy||8 (3)||11 (2)|
| Logistical||1 (0)||1 (0)|
| Reason unknown||30 (10)||55 (10)|
|Study protocol|| || |
| Physician decision/ adverse event||7 (2)||9 (2)|
| Study violation||8 (3)||11 (2)|
| Noncompliance||16 (5)||24 (4)|
|Not documented||31 (10)||77 (14)|
Patient Characteristics Associated With Attrition
Table 4 shows the univariate and multivariate analyses to identify factors predictive of attrition. We found that a higher intensity of fatigue was associated with higher rates of dropout before the primary endpoint. The results of the current study also revealed that Hispanic race, higher educational level, non-Christian religious affiliation, and a higher intensity of dyspnea and fatigue were associated with dropout before the end of the study. When we repeated the above analyses but without the 2 inpatient studies, the findings were similar, with the exception that an advanced educational level was also associated with attrition before the primary endpoint.
Table 4. Patient Factors Associated With Attrition Before the Primary Endpoint and Before the End of the Study
|Age||1214||.06||1.01 (1.00–1.02)||.10||1214||.1||1.01 (1.00–1.02)||.11|
|Female sex||178 (57)||507 (56)||.38|| || ||291 (54)||394 (58)||.45|| || |
|Race|| || ||.17|| || || || ||.008d|| || |
| Asian||20 (6)||49 (5)|| || || ||28 (5)||41 (6)|| ||1.74 (0.96–3.16)||.07|
| Black||51 (16)||150 (17)|| || || ||81 (15)||120 (18)|| ||0.85 (0.59–1.23)||.39|
| Hispanic||52 (17)||108 (12)|| || || ||87 (16)||73 (11)|| ||1.87 (1.26–2.76)||.002|
| White||188 (60)||596 (66)|| || || ||339 (63)||445 (66)|| ||1||Reference|
|Married||190 (61)||577 (64)||.63|| || ||331 (62)||436 (64)||.82|| || |
|Educational level|| || ||.10|| || || || ||.05|| || |
| High school or less||163 (52)||461 (51)|| || || ||263 (49)||361 (53)|| ||0.67 (0.46–0.97)||.04|
| College||84 (27)||319 (35)|| || || ||170 (32)||233 (34)|| ||0.64 (0.43–0.95)||.03|
| Advanced degree||64 (21)||123 (14)|| || || ||102 (19)||85 (13)|| ||1||Reference|
|Cancer diagnosis||311 (100)||903 (100)||.63|| || ||535 (100)||679 (100)||.62|| || |
|ECOG PS|| || ||.19|| || || || ||.045|| || |
| 0–1||128 (41)||378 (42)|| || || ||211 (39)||295 (43)|| ||0.79 (0.56–1.10)||.16|
| 2||68 (22)||224 (25)|| || || ||118 (22)||174 (26)|| ||0.75 (0.51–1.10)||.15|
| 3–4||115 (37)||301 (33)|| || || ||206 (39)||210 (31)|| ||1||Reference|
|ESAS Anxiety||1214||.90|| || ||1214||.53|| || |
|ESAS Appetite||1214||.01||1.04 (0.99–1.09)||.11||1214||.03||1.02 (0.97–1.07)||.38|
|ESAS Depression||1214||.78|| || ||1214||.69|| || |
|ESAS Drowsiness||1214||.65|| || ||1214||.69|| || |
|ESAS Fatigue||1214||.0001||1.10 (1.03–1.17)||.01||1214||.0009||1.08 (1.02–1.15)||.01|
|ESAS Nausea||1214||.32|| || ||1214||.10|| || |
|ESAS Pain||1214||.04||1.03 (0.98–1.09)||.26||1214||.06||1.01 (0.96–1.07)||.59|
|ESAS Sleep||1214||.21|| || ||1214||.09||1.02 (0.98–1.07)||.31|
|ESAS Dyspnea||1214||.002||1.05 (1.00–1.10)||.058||1214||.0002||1.06 (1.01–1.11)||.009|
|ESAS Well-being||1214||.12|| || ||1214||.09||0.99 (0.93–1.04)||.61|
Clinical Trial Characteristics Associated With Attrition
At the study level, we found a longer study duration was associated with attrition before the primary endpoint (Spearman correlation, 0.49; P = .04) and the end of the study (Spearman correlation, 0.59; P = .01). Outpatient studies were also more likely to experience patient dropout before the end of the study (47% vs 6% for inpatient studies; P = .05) (Table 5).
Table 5. Study Factors Associated With Attrition Before Primary Endpoint and Before the End of the Study
|Anorexia or cachexia clinical trial||39 (34–41)||26 (3–63)||.14||53 (46–74)||35.7 (3–86)||.21|
|External funding||26 (4–56)||34 (3–63)||.51||36 (4–86)||46 (3–75)||.76|
|Fatigue clinical trial||38 (3–50)||50 (4–86)||.08||38 (3–50)||50 (4–86)||.31|
|Multicenter study||31 (3–63)||6 (4–8)||.19||25 (4–69)||48 (3–86)||.19|
|Outpatient study||31 (3–63)||6 (4–8)||.07||48 (3–86)b||6 (4–8)||.049|
|Randomized controlled trial||30 (4–60)||26 (3–63)||.77||46 (4–86)||26 (3–75)||.52|
Although all 18 clinical trials included in the current study were designed and conducted by an experienced team of researchers and the study criteria were devised to minimize attrition, 1 in 4 patients dropped out before the primary endpoint, and 1 in 2 patients dropped out before the end of the study. The attrition rate varied widely among studies. Hispanic race, higher educational level, non-Christian religious affiliation, and a high symptom burden at the time of enrollment were found to be associated with a higher risk of dropping out. We believe these findings have implications for future study design, including sample size estimation and measures to minimize patient dropout.
To our knowledge, this is the most comprehensive study published to date examining the issue of attrition in an aggregate of symptom control clinical trials. The rate of attrition noted in the current study is generally consistent with those reported in the literature. Oldervoll et al reported that 46% of cancer patients dropped out of a phase 2 feasibility study of exercise in the palliative setting.21 In a study that was funded by the National Cancer Institute, McMillan et al found that only 38% of patients had complete data available at the time of follow-up.22 A study of 40 randomized controlled trials regarding cognitive behavioral interventions for pediatric chronic pain found a mean attrition rate of 20% (range, 0%-54%) for initial follow-up and 32% (range, 0%-59%) for extended follow-up.23 It is interesting to note that the rate of attrition in our symptom control trials is also comparable to that of other palliative care studies on health services research. A review of methodological issues in effectiveness research on palliative cancer care found high dropout rates in the clinical trials that ranged from 34% to 80%.7 More recently, in a systematic review evaluating the effectiveness of specialist palliative care, Zimmermann et al reported that the median rate of loss to follow-up for quality of life and satisfaction was 40% (range, 3%-92%) among 20 studies.24
The major reasons for attrition were patient deterioration and patient decision to withdraw from the study, which accounted for approximately 80% of the dropouts. This is not surprising given that patients enrolled in palliative care clinical trials generally have a poor performance status and short life expectancy. Many acute complications occur in the last few weeks/months of life, resulting in acute deterioration, worsening distress, hospitalizations, and death, thereby affecting a patient's ability and willingness to continue with the study.25 Importantly, others have also reported that death, physical decline, and emotional distress were major contributors to patient dropout.9, 22, 26 We found that one of the key reasons for patients to withdraw from palliative care studies is increasing symptom distress. Further studies are required to determine whether this is because of the natural progression of disease, adverse effects related to the study intervention, or the inability of the study intervention to control the targeted symptom. It is interesting to note that a few patients dropped out because of safety concerns or protocol violations, or were lost to follow-up. This may be explained by the attention paid to study design and careful follow-up by our research team.
As demonstrated in the current study, attrition can have a major impact on the quality of the study. McWhinney et al initiated a randomized controlled trial to evaluate a palliative care home support team, but had to terminate the study early because of an attrition rate of 36%.9 The identification of risk factors for dropout may allow us to design studies with minimal attrition. We found that Hispanic race, non-Christian religious affiliation, and a higher educational level were associated with higher dropout rates on multivariate analysis. Others have also identified minority race as a contributor to attrition.27, 28 This may be related to distrust of the medical system, language barriers, and lack of resources (eg, transportation to hospital), making it less appealing for members of minority groups to remain on the study.29-31 Further research is needed to determine why patients with lower levels of education were more likely to remain on the clinical trials in the current study. Furthermore, we found that high levels of fatigue and dyspnea and a poor ECOG performance status were predictors of attrition. This is ironic given that the primary concern for these clinical trials, namely symptom burden and function, are the very contributors to attrition. This highlights the unique challenges in conducting clinical trials in the palliative care setting, particularly for dyspnea and fatigue studies. We also found that inpatient studies had the best overall retention rates, which may be explained by the short duration and direct supervision of care in these trials.
How can we minimize the rate of patient dropout from symptom research clinical trials? We recommend that all research protocols should keep the study as short as possible, minimize the study burden, and incorporate close monitoring and support for the patients. This is particularly important for fatigue and dyspnea studies. For explanatory trials, investigators may choose to enroll patients who are most likely to complete the study, such as hospitalized patients, non-minority individuals, those with a longer life expectancy, and those with a better performance status.32 Moreover, we need to recognize that high attrition may be inevitable in symptom research clinical trials, particularly phase 3 trials of a pragmatic nature, research on dyspnea and fatigue, and studies involving minority and end-of-life (ie, weeks to days of survival) populations. Findings from symptom control trials conducted in patients with early cancer may not be generalizable to those patients with advanced disease because of their unique needs. Thus, it is imperative to conduct studies involving patients who are frail and symptomatic, despite a higher expected attrition rate. Funding proposals and research protocols should plan to enroll a higher number of patients such that an adequate sample size can be achieved for the primary outcome.
Only 2 of the 18 trials reviewed in the current study enrolled the planned number of patients.16, 17 The small sample size precludes detailed statistical analysis. Funding, study design, and committed research staff are likely to be important contributors to successful enrollment. Further research is needed to identify factors associated with study completion.
The current study has several limitations. First, the clinical trials were all conducted by a single research group at a tertiary care cancer center. The findings of this study may not be generalizable to other settings. Second, although data were collected prospectively, some variables such as ECOG performance status and education were not routinely collected in all clinical trials. Imputation techniques were used to maximize the data available for analysis. Future studies should routinely collect data concerning symptom batteries, performance status, and cognitive status to facilitate the interpretation of findings and comparison of patient populations. Third, we did not include several variables such as patient interest, history of participation, and travel distance, which could potentially account for attrition. Fourth, the number of clinical trials was small, and thus did not allow for a detail analysis of study-related factors associated with attrition. Further studies are needed to examine this issue in more detail. Finally, patient deterioration leading to dropout may be related to cancer progression, adverse effects of treatment, and/or the failure of supportive therapies to control symptoms, and it is often difficult to distinguish between these possibilities.
Attrition was found to be high among our cohort of symptom control clinical trials. Patient deterioration and withdrawal were the major reasons for dropouts. Various patient characteristics, poor ECOG performance status, and a high symptom burden were identified as predictors of attrition. To safeguard the scientific integrity of palliative care clinical trials, investigators need to routinely anticipate attrition and incorporate various measures to minimize patient dropout during trial design.
Supported in part by National Institutes of Health grants RO1NR010162-01A1, RO1CA122292-01, and RO1CA124481-01 (to Dr. Bruera). Also supported by The University of Texas MD Anderson Cancer Center Support Grant (CA 016672) and an institutional startup grant (#18075582) (to Dr. Hui). The sponsor of the study had no role in the study design, data collection, analysis, interpretation, or writing of the report.
CONFLICT OF INTEREST DISCLOSURES
The authors made no disclosures.