Primary care physician documentation in oncology electronic health records: prognostic marker or immortal-time bias?

BACKGROUND: Structured administrative fields in oncology electronic health records (EHRs) are used in studies but their interpretation may depend on when they are completed. We assessed whether dated primary care physician (PCP) documentation was associated with survival after accounting for documentation timing. MATERIALS AND METHODS: We conducted a retrospective cohort study using routinely collected oncology EHR data from a single French comprehensive cancer centre (ConSoRe platform), including patients treated for cancer between 2015 and 2017. Survival was analysed from cancer care initiation to death from any cause or administrative end of follow-up; no patients were censored before 12 months. Conventional analyses compared patients with or without an exploitable dated PCP declaration using inverse probability of treatment weighting (IPTW) targeting the average treatment effect based on age, sex, metastatic status, and tumour location. Temporally aligned analyses included baseline-available exposure definitions, a time-dependent Cox model, and a 6-month landmark analysis. RESULTS: Among the 15 216 patients, 10 418 had a dated PCP declaration (68.5%), 4757 had a declared PCP but no exploitable date (31.2%), and 41 had no PCP declared (0.3%). During the median follow-up of 102.1 months, 7175 deaths occurred. In conventional IPTW analysis, dated PCP documentation was associated with improved survival [hazard ratio (HR) 0.86, 95% confidence interval (CI) 0.82-0.90, P < 0.001]. Patients with declarations recorded after 6 months had the most favourable survival, consistent with immortal-time bias. In the time-dependent Cox model, dated documentation was not protective (HR 1.15, 95% CI 1.10-1.21, P < 0.001). In the 6-month landmark analysis, unadjusted postlandmark survival did not differ significantly (log-rank P = 0.140), whereas the adjusted landmark Cox model showed a small association in the opposite direction to the conventional analysis (HR 0.94, 95% CI 0.89-0.99, P = 0.011). CONCLUSIONS: The apparent survival benefit associated with dated PCP documentation was explained by documentation timing rather than a robust prognostic effect. Time-dependent EHR fields require temporally aligned analyses in real-world oncology research.