Original ArticleIn a randomized controlled trial, missing data led to biased results regarding anxiety
Introduction
The hypothesis that psychosocial intervention could affect the well-being of cancer patients has been subject to research for some 30 years [1]. The randomization used in clinical trials protects against bias in treatment allocation, but missing values may still lead to severe bias in a randomized trial, or in any other intervention trial [2], [3], [4], [5], [6], [7].
Besides death, the main reason for missing observations is nonresponse by choice: that is, refusal to participate in one or more follow-up assessments. Given the obvious large psychosocial differences between nonresponse by choice and nonresponse by death, the associated factors are bound to differ. Insight into the mechanisms will be severely weakened if nonresponse and death are mixed. Furthermore, differential associations for the two randomization groups may be more likely for nonresponse. A low level of well-being or a high level of distress (singly or in combination) could be associated with unwillingness among all participants to take the trouble to participate in an interview. A falsely lower level of anxiety and depression will be observed in the control group if, for example, the distressed patients in the intervention group feel obliged to participate in the interview anyway, but the patients in the control group feel free to decline participation in times of distress. Such a differential association between well-being and nonresponse could diminish or even reverse the estimated effect of the intervention on the well-being variables. It is therefore important for any clinical trial of well-being to check that differential nonresponse has not occurred.
The problem of missing data has been dealt with in the statistical literature, with distinctions drawn among different missing data mechanisms: missing completely at random, missing at random, and informative missing. Several papers report different ways to distinguish among these missing data mechanisms [5], [6], [8], [9], [10], [11], [12], [13], [14], [15]. The method for dealing with missing data depends on the missing data mechanism and the missing data pattern (drop-outs or missing intermittently), as well as the type of the outcome variable (continuous, ordinal, binary, or failure time data) [10], [11], [13], [15], [16], [17], [18], [19], [20], [21], [22]. A few papers focus on quality of life data [22], [23], [24], but psychosocial intervention studies reporting on missing data usually consider sociodemographic variables only [25], [26], [27], [28]. Other aspects of well-being may, however, be very important factors in relation to the missing data mechanism.
Based on a randomized trial of psychosocial intervention, this study investigates the associations between well-being and missing observations at the next follow-up. It is hypothesized that distress increases the probability of nonresponse, but that this may be less pronounced in the intervention group. Furthermore, another study has found an association between distress and imminent death [29], and we wanted to investigate whether this association could be explained by concurrent physical health as measured by physical functioning and global health status and quality of life.
Section snippets
Materials and methods
In the period September 1996 to May 1999, a total of 249 newly diagnosed Danish colorectal cancer patients were included in a randomized study to investigate the psychological effects of home visits (the INCA Project). Patients were aged 18 or older and had undergone abdominal surgery. At baseline, patients filled in a questionnaire concerning health habits and sociodemographic data; they were subsequently randomized into a control group or an intervention group. Patients in the intervention
Results
At baseline, we observed no significant differences between the two randomization groups on sociodemographic or clinical variables (Table 1).
At the first follow-up interview 3 months after discharge, 4% of the patients in the intervention group and 2% of the patients in the control group had died. These proportions had risen to 34% and 29%, respectively, at the final follow-up interview 2 years after discharge (Table 2). The reasons for nonresponse consisted of nonresponse by choice (refusal,
Discussion
In this analysis of participation status at the succeeding follow-up categorized into participation, death and nonresponse for reasons other than death, we found a differential association between anxiety and nonresponse in the two randomization groups, but no differential associations were seen for death. The probability of nonresponse decreased with increasing anxiety score in the intervention group, but increased with increasing anxiety score in the control group. Furthermore, we found that
Conclusion
In the present study, we found differential nonresponse from reasons other than death related to anxiety in a randomized clinical trial of the effect of psychosocial intervention on well-being. Bias due to differential nonresponse may severely invalidate the results regarding the effect of this kind of intervention, and we therefore strongly recommend that psychosocial intervention studies include an analysis of the pattern of nonresponse before concluding on the effect of the intervention. In
References (38)
- et al.
Mind and cancer: does psychosocial intervention improve survival and psychological well-being?
Eur J Cancer
(2002) Meta-analysis adjusting for compliance: the example of screening for beast cancer
J Clin Epidemiol
(1992)Statistical considerations in the intention-to-treat principle
Control Clin Trials
(2000)- et al.
The art of quality assessment of RCTs included in systematic reviews
J Clin Epidemiol
(2001) - et al.
Sample size slippages in randomised trials: exclusions and the lost and wayward
Lancet
(2002) - et al.
Intention-to-treat approach to data from randomized controlled trials: a sensitivity analysis
J Clin Epidemiol
(2003) - et al.
Intent-to-treat analysis for longitudinal clinical trials: coping with the challenge of missing values
J Psychiatric Res
(1999) - et al.
Social relations: network, support and relational strain
Soc Sci Med
(1999) - et al.
The construction and testing of the EORTC colorectal cancer-specific quality of life questionnaire module (QLQ-CR38)
Eur J Cancer
(1999) - et al.
Impact of missing data due to selective dropouts in cohort studies and clinical trials
Epidemiology
(2002)
A test of the missing data mechanism for repeated categorical data
Biometrics
A test of missing completely at random for longitudinal data with missing observations
Stat Med
Model-based inference for categorical survey data subject to non-ignorable non-response
J R Stat Soc [Ser B]
Analysis of longitudinal data with non-ignorable non-monotone missing values
J R Stat Soc Ser C Appl Stat
Simple pattern-mixture models for longitudinal data with missing observations: analysis of urinary incontinence data
Stat Med
Estimation and comparison of rates of change in longitudinal studies with informative drop-outs
Stat Med
A test of missing completely at random for generalised estimating equations with missing data
Biometrika
Influence of human immunodeficiency virus infection on neurological impairment: an analysis of longitudinal binary data with informative drop-out
J R Stat Soc Ser C Appl Stat
Statistical analysis with missing data
Cited by (24)
Monte Carlo simulation of the cost-effectiveness of sample size maintenance programs revealed the need to consider substitution sampling
2012, Journal of Clinical EpidemiologyCitation Excerpt :This threat has the potential to undermine the interpretability or credibility of results [7–9]. Study results may be biased and have reduced validity because of differential nonresponse between comparison groups or by differences between those who attrite and those that continue to participate [10–14]. In addition to these potential consequences, the reduction of sample size by subject loss throughout the duration of a study also has a negative impact on statistical power [15,16].
Characteristics of nonparticipants differed based on reason for nonparticipation: a study involving the chronically ill
2008, Journal of Clinical EpidemiologyCitation Excerpt :Comparisons with the literature for results based on reason for nonparticipation are difficult due to the variation in classification and inclusion criteria. However, our findings are consistent with others that have found that those deceased were more likely to be older [11], male [4,11], not married [6], and experiencing ill health [6]. Those not participating due to refusal were more likely to be female [11].
Discordance between reported intention-to-treat and per protocol analyses
2007, Journal of Clinical EpidemiologyCitation Excerpt :If the dropouts in a trial are similar to future dropouts, and given that there is a good definition of the studied, treated, and sick populations [12], it can be argued that a valid study-based ITT analysis will adequately address use effectiveness. On the other hand, as dropout may be related to outcome [13] and it may have different cause in each treatment arm [14], the PP estimate that excludes protocol deviations will be biased [6,15,16], especially when there is a large percentage of dropout [17]. Furthermore, compliance can interact with treatment, resulting in better results for compliers in the active group but just the opposite (better for noncompliers) in the control group [10].
Sickness absence, employment history, and high-risk employees: A 10-year longitudinal study
2010, Journal of Occupational and Environmental Medicine