Practical Fraud Detection and Prevention in Incentivized Online Surveys: Lessons from the ADOPT Study (Preprint)

Vollständiger Abstract

Worum geht es in dieser Arbeit?

BACKGROUND Digital surveys are increasingly integrated into clinical and public health research to capture patient-reported outcomes. However, concerns about fraudulent or duplicate responses threaten data integrity. Most literature on survey fraud focuses on open-access online recruitment, where bot-generated or anonymous entries are common, but far less is known about fraud patterns in clinic-linked, incentive-based surveys. Evaluations of real-world implementation of fraud deterrence strategies remain limited. OBJECTIVE This study evaluated whether implementing enhanced fraud deterrence procedures in an incentive-based postoperative survey reduced the prevalence of potentially fraudulent responses. Secondarily, the study evaluated which indicators were most frequently triggered before and after implementation. METHODS This evaluation was conducted within the Alternatives to Dental Opioid after Tooth Extraction (ADOPT) study. Eligible patients aged 12-25 years were recruited through QR-coded, clinic-distributed flyers and invitation cards. Participants completed a screening survey followed by an incentivized postoperative survey between days 6-10 after tooth extraction. A mid-study protocol modification introduced enhanced fraud deterrence measures including required phone numbers, prohibiting email invitations, confirming date of birth across projects, and shifting to a manually reviewed participant list with SMS text message invitations. Responses were categorized into “control” (before modification) or intervention (after modification). An eight-item fraud scoring system assessing completion time outliers, late-night submissions, repeated screeners, unusual email formats, repeated phone numbers, impossible logic, blank recruitment source, and illogical response patterns was used to classify responses as potentially fraudulent (≥2 indicators). Sensitivity analyses evaluated thresholds from 1-3 indicators. RESULTS A total of 573 were included (control n=122; intervention n=451). The overall prevalence of potentially fraudulent responses was low (6.8%) and did not differ significantly between cohorts (7.4% vs. 6.7%, p=.778). Certain indicators shifted between cohorts: blank recruitment source declined (12.3% to 4.9%, p=.003), and repeated phone numbers increased (3.0% to 25.1%, p