Between-assessor and temporal heterogeneity in routinely recorded postoperative symptom fields: a five-year retrospective observational study of 121,755 nursing assessments

Abstract Background Routine postoperative patient-controlled analgesia (PCA) follow-up records may be reused for audit or benchmarking, but recorded symptom rates reflect patient, clinical, temporal and documentation factors. We quantified between-assessor and temporal heterogeneity and examined what remained estimable after available adjustment. Methods After removing 5,519 excess completely identical rows, we analysed 121,755 PCA follow-up assessments from 45,398 surgical patients (January 2020–April 2025). Non-identical same-patient/same-time rows were retained and disclosed. American Society of Anesthesiologists (ASA) physical status was linked only by patient plus exact encounter/start time. The primary cohort contained 120,680 assessments from 45,177 patients and 15 assessors. Primary-cohort M0–M4 logistic mixed models addressed repeated patients, year, ward and available case mix. After the prespecified optimiser and fallback attempts for sedation M4 failed, the corresponding author approved a post hoc amendment to the analysis plan before the remaining restricted-cohort models were run. Those cohorts were summarised descriptively with Wilson 95% confidence intervals. Results Corrected assessor-specific rates ranged from 0.14% to 13.19% for nausea/vomiting code ≥ 1, 0.10% to 16.43% for dizziness code ≥ 1 and 0% to 7.78% for documented sedation score ≥ 1. Assessor-only variance partition coefficients (VPCs) were 28.1%, 42.5% and 67.6%, respectively. Of 14 primary-cohort models with numeric result records, 9 met the prespecified numerical acceptance criteria and 5 were not estimable; 3 accepted fits were singular. M3 results were < 0.001% (median odds ratio [MOR] 1.00; singular), < 0.01% (MOR 1.02; singular) and not estimable, respectively. Sedation M4 remained not estimable after the prespecified fallback attempts. The six restricted models were not run after the analysis-plan amendment; descriptive rates are reported without adjusted clustering estimates. Conclusions Corrected local symptom fields showed substantial recorded-rate heterogeneity, but the adjusted ladder was incompletely estimable and did not identify its source. The results do not support nurse ranking or inference about true symptom incidence. Prospective validation is required before benchmarking.