BackgroundPost-COVID conditions (PCC) present clinicians with significant challenges due to their variable presentation.ObjectiveTo characterize patterns of PCC diagnosis in generalist primary care settings.DesignRetrospective observational studySetting519 primary care clinics around the United States who were in the American Family Cohort registry between October 1, 2021 and November 1, 2023.Patients6,116 with diagnostic code for PCC; 5,020 with PCC and COVID-19MeasurementsTime between COVID-19 and PCC (U09.9) diagnostic codes; count of patients with PCC diagnostic codes per clinician; patient-specific probability of PCC diagnostic code estimated by a tree-based machine learning model trained on clinician and specific practice visited, patient demographics, and other diagnoses; performance of a natural language classifier trained on notes from 5,000 patients annotated by two physicians to indicate probable PCC.ResultsOf patients with diagnostic codes for PCC and COVID-19, 43.0% were diagnosed with PCC less than 4 weeks after initial recorded COVID-19 diagnostic code. Six clinicians (out of 3,845 total) made 15.4% of all PCC diagnoses. The high-performing (F1: 0.98) tree-based model showed that patient demographics, practice visited, clinician visited, and calendar date of visit were more predictive of PCC diagnostic code than any symptom. Inter-rater agreement on PCC diagnosis was moderate (Cohen’s kappa: 0.60), and performance of the natural language classifiers was poor (best F1: 0.54).LimitationsCannot validate date of COVID-19 diagnosis, as it may not reflect when disease began and could have been coded retrospectively. Few options for medically focused language models.ConclusionWe identified multiple sources of heterogeneity in the documentation of PCC diagnostic codes in primary care practices after introduction of ICD-10 codes for PCC, which has created challenges for public health surveillance.Funding SourceUS CDC