Presentation (CTI)

OBJECTIVES: Nonresponse to patient-reported outcome measures is common and can vary over time in real-world longitudinal research. Such missingness can introduce bias that not only compromises the generalizability of study findings but may also lead to misinterpretation. In a case study examining the association between symptom assessment completion and chemotherapy (CTX) discontinuation, we proposed using inverse probability weighting (IPW) to address the time-dependent bias introduced by nonresponse.
METHODS: Data was derived from adult patients with gastrointestinal (GI) cancers who received CTX at Mass General Brigham between 01/2019 and 01/2024. The Patient-Reported Outcome version of the Common Terminology Criteria for Adverse Events assessed 12 symptoms at CTX initiation and at days 30, 60, and 90. All patients completed baseline assessments. The primary outcome was discontinuation due to toxicity within 90 days of initiation. Using the Fine-Gray model to account for competing risks of death and progression, we examined the association between discontinuation and number of completed assessments over time. To address the time dependent nonresponse bias associated with disease severity, treatment, and prior completion, we constructed IP weights for each follow-up time point.
RESULTS: Among 1,178 patients, 67% completed assessments at baseline and during follow-up. The unweighted model without adjustment for nonresponse bias found significant relationship between number of symptom assessment completion and discontinuation (SHR = 0.49; 95% CI: 0.25-0.95). In the weighted model accounting for nonresponse bias, completing additional symptom assessments after CTX initiation was associated with a lower risk of discontinuation; however, this relationship was not statistically significant (SHR = 0.70; 95% CI: 0.44-1.12).
CONCLUSIONS: Our findings highlight the importance of recognizing and addressing time-dependent nonresponse bias in real-world longitudinal research using patient-reported outcome measures. We advocate for the application of appropriate methods, such as IPW, to manage such bias and improve the validity of study findings.