To develop an algorithm to identify pelvic inflammatory disease (PID) episodes based on diagnostic codes, estimate their frequency and determine the proportion related to Chlamydia trachomatis infection.

Hospital visits by women aged 18–45 years presenting with one of 31 diagnostic codes potentially associated with PID between July 2017 and December 2019 were extracted from the electronic medical record (EMR) warehouse of 39 public hospitals in the Paris area, France (EDS-APHP).

1956 patients totalling 2191 visits were included.

Natural language processing (NLP) methods were developed and validated against expert reading to identify PID episodes from textual medical reports. The resulting classification served as a reference to select relevant diagnostic codes for constructing a PID identification algorithm.

The final algorithm included 10 diagnostic codes, with 3 codes searched in principal or associated diagnoses and 7 codes in principal diagnosis only. The algorithm’s performance on 1732 NLP-classified visits was a 0.80 recall, a 0.74 precision and a 0.77 F1-score. On a set of 93 expert-classified visits, these performance metrics were 0.76, 0.82 and 0.79, respectively. The algorithm identified 901 PID episodes in 880 patients during the study period. C. trachomatis infection was detected in 9.9% of them.

Exploiting EMR data enabled the development of an algorithm with satisfactory performance for detecting PID episodes based solely on diagnostic codes. This algorithm will be useful to identify PIDs among hospitalisations from large-scale healthcare claims databases lacking medical reports, which will contribute to assess the burden of C. trachomatis infections and their complications.