Self-selection and decision to participate in clinical trials: evidence from screening trials
This project will use data from National Lung Screening Trials (NLST) and Health and Retirement Survey (HRS) to address why enrollees self-select into the clinical trials, compared with non-participants filtered from HRS according to the eligibility criteria. Specifically, under this screening trial, the participants face the tradeoff between the potential benefits and costs. The potential benefits include detecting lung diseases and other diseases earlier, which leads to a higher chance of surviving from the disease, satisfaction of being part of scientific research, and possibly money paid by the conductor of clinical trials. The potential costs include time to participate, a more frequent exposure under radiation, a possibly lower disease detection probability if assigned to the arm with chest radiography, discomfort from procedures, and negative psychological impact due to false alarms (Liran et al. 2020). Before the clinical trials, participants must be well-informed about the potential benefits and costs. But how they value the benefits and what their participation cost is are yet to be known.
The question on why they participate in screening trials also touches on the relationship between self-selection into screening and the phenomenon of overdiagnosis. Patz et al. (2014) concluded that more than 18% of all lung cancer diagnoses in NLST seem to be overdiagnosed. However, as Kevin et al. (2015) argued, the estimation in Patz has a potential bias because of the problems embedded in the design of the screening trials. This project will address the role played by self-selection for the bias. By understanding participants’ incentives, we can identify how much bias is generated by their self-selection. Then, more precise estimates can explain the overdiagnosis phenomenon from the patients’ perspective.
This project will adopt the choice-based sampling model to explain participants’ value of benefits and quantify their participation cost. The estimators will be applied to evaluate the effectiveness of health guidelines on screening for lung cancer considering patients’ interests. Also, the model will quantify the uncertainties embedded in medical research and help evaluate the value and cost to do clinical trials. Moreover, the model can be generalized to scenarios when designing the regulation involving voluntary participation.
Current literature has addressed the problems of extrapolating from the study sample of clinical trials to patient populations (Manski 2017; Kevin et al. 2015; Fleming and Demets, 1996). This project will approach this extrapolation from the participants’ self-selection decision and the comparison with non-participants. Participants do not share all characteristics with the population partly because of their self-selection incentives. Identifying participants’ incentives and comparing them with non-participants can better our understanding of the difference between the study sample and the patient population to improve the extrapolated results. Another group of literature focuses on the impact of health guidelines on selection (Emily 2020; Liran et al. 2020; Liran et al. 2020, working paper; Kowalski 2021, working paper). Many of them focus on guidelines on mammograms. With NLST data, some medical researchers have approached the overdiagnosis problem in low-dose helical computed tomography (CT) (Patz et al. 2014; Sandra 2020). But this project will consider participants’ choices to explain the rationale behind the screening and possible overdiagnosis from CT and chest radiography arms. Considering two arms in the screening trials can improve the overall overdiagnosis estimation. The long-term goal is to apply the model to evaluate whether the health guidelines on CT for the high-risk group of people are appropriate and to generalize it under different scenarios. The specific aims are as follows:
Aim 1: Identify participants’ incentives from the hypothesis by constructing participants’ decision model: Participants decide to participate in the screening trial because their benefits outweigh participation costs.
Aim 2: Identify participants’ self-selection: whether people are more likely to participate in the clinical trials when they are at higher risk and value the benefits more than non-participants. Non-participants are filtered from HRS data. The filter corresponds to the eligibility criteria of the screening trial.
Aim 3: Quantify the participation cost. Participation costs can be identified by the difference in the frequency of screening between participants and non-participants and the false positives inferred from data. In addition, the difference in the length of the time from diagnosis to death between the participants assigned to do chest radiography and non-participants can identify the participation cost associated with uncertainty for the screening trial.
Aim 4: Identify how much self-selection bias accounts for the bias of overdiagnosis estimates with NLST data by comparing non-participants’ disease history.
Steven Stern, Stony Brook University