Evaluation of Psychiatric Diagnoses as a Source of Bias in Prognostication and Decision-Making for Palliative Radiotherapy

Purpose/Objective(s)

Up to half of patients with advanced cancer may meet diagnostic criteria for a psychiatric disorder. Yet, the impact of psychiatric diagnosis and/or medication prescription (psych Dx/Rx) on decision-making and outcomes for radiotherapy (RT) are not well-characterized. Statistical models and decision tools can be skewed when built from source data that has been influenced by implicit and explicit biases in medical diagnoses and treatment. We investigate the impact of psych Dx/Rx on palliative RT outcomes and decisions to ascertain whether this may serve as a source of bias in prognostication and decision-making in this patient population.

Materials/Methods

Retrospective review identified patients treated with conventional external beam (cEBRT) or stereotactic body RT (SBRT) for bone metastases at 3 radiation oncology facilities from 5/2013-12/2019. Prevalence of psych Dx/Rx was collected, along with 27 other patient, disease, and treatment characteristics that were identified as collectively predictive of survival using the validated Bone Metastases Ensemble Trees for Survival (BMETS) machine-learning model and decision support platform. RT details were documented. Overall survival from consultation to death or last follow up was estimated using the Kaplan-Meier method. Uni- and multivariable regressions evaluated associations between psych Dx/Rx, survival, and RT fractionation and modality. Reported statistics are p<.05.

Results

346 treated patients were evaluated, including 178 patients with cEBRT and 168 patients with SBRT (median 10 vs. 3 fractions). Psych Dx/Rx was more prevalent among patients treated with cEBRT vs. SBRT (40% vs. 28.6%). On univariate analysis, odds of psych Dx/Rx was significantly higher for patients requiring opiate analgesics, females, white (vs. black) race, and breast (vs. prostate) cancer (OR 2.0, 3.7, 4.3, 5.6). Psych Dx/Rx was associated with use of fewer fractions for cEBRT (β coefficient -1.0) but not SBRT. On multivariable analyses for the cEBRT cohort controlling for other univariable associations, use of fewer fractions remained significantly associated with psych Dx/Rx (β coefficient -1.1). While overall survival was significantly longer for patients treated with SBRT (median 3.3 vs. 0.8 years for cEBRT), psych Dx/Rx was not associated with survival or strongly linked to other predictors of advanced disease for either RT cohort.

Conclusion

Psych Dx/Rx does not appear to impact survival for patients undergoing palliative bone RT, arguing against its direct role as a source of bias in survival models like BMETS. However, given that psych Dx/Rx is associated with lower use of SBRT and foreshortened cEBRT regimens, this variable may serve a source of bias in treatment selection, albeit perhaps concordant with the patient's best interest. This highlights the importance of enhanced shared decision-making in this potentially vulnerable patient population.