International Classification of Diseases codes do not capture all cases of hidradenitis suppurativa in the electronic health record: a retrospective cohort

Dear Editor,

Hidradenitis suppurativa (HS) is a chronic inflammatory skin disorder of the hair follicle. Previous studies show that International Classification of Diseases (ICD) codes can identify cases of HS with high positive predictive value (PPV).^1–3 HS frequently has diagnostic delays of 5 years or more.⁴ We hypothesized that using ICD codes alone to identify cases of HS would miss many true cases, which could bias associations in electronic health record (EHR) cohorts. We therefore conducted this study to estimate a more accurate sensitivity.

This study was approved by the institutional review board at Vanderbilt University Medical Center as nonhuman research, IRB #200335. Following approval, we identified potential patients with HS in the Synthetic Derivative (SD). The SD is a deidentified mirror image of the EHR that contains information on demographics, diagnosis and procedure codes, clinical notes and other details. We identified patients with at least one ICD-9 code 705.83, ICD-10 code L73.2, or Current Procedural Terminology (CPT) code for HS surgery (11450-1, 11462-3, 11470-1); or any patient with a keyword of ‘hidradenitis’ or ‘hydradenitis’ excluding ‘neutrophilic eccrine hidradenitis’. Data were available for dates between October 1989 and February 2020. Charts were manually reviewed for HS diagnosis, including instances where an official diagnosis had not been made but there was sufficient information provided to make a clinical diagnosis. Equivocal cases were considered noncases. Chart review and data analyses were conducted from March 2020 to November 2021.

There were 3131 patients with at least one ICD code for HS, and 5251 with a CPT code or mention of ‘hidradenitis’ or ‘hydradenitis’ in the chart, for a total of 5735 potential cases. Of these, a random sample of 1650 were reviewed for case status. Following review, routine SD maintenance identified twelve patient records that had duplicates and merged them into single records per patient, resulting in a final sample size of 1638 charts. Differences between groups were tested using the Student’s t-test or χ²-test for continuous or categorical variables. There were no significant differences between the reviewed and unreviewed charts by sex (74·5% female vs. 75·1%), race (59·6% white vs. 58·9%), mean (SD) age [44·3 (16·3) years vs. 44·2 (16·3) years], mean (SD) number of HS ICD codes [4·1 (23·8) vs. 4·0 (16·6)] or mean (SD) number of HS-related CPT codes [0·8 (3·2) vs. 0·8 (2·9)] (P > 0·05 for each). Patients with reviewed charts were less likely to have received biologic treatment or to have an HS keyword (4·9% vs. 6·5%, P = 0·02). The mean duration of follow-up was significantly greater for cases than for noncases (2·3 years vs. 1·7 years, P = 0·03).

Most of the patients whose charts were reviewed met the clinical diagnosis for HS (n = 1316, 80·3%), while 322 (19·7%) did not. Of the noncases of HS, 295 (91·6%) were identified by keyword search criteria without an HS ICD code, although 27 of these noncases had at least one ICD code for HS. Most noncases with an ICD code had no notes available for review (n = 16) or had only an isolated case of folliculitis (n = 6). Of the cases, only 834 (63·4%) had an ICD code specific for HS, while 482 (36·6%) were identified through HS keywords. Cases were more likely to be female (76·3%, P < 0·001), but female cases were less frequently given an HS ICD code than male cases (71·8% vs. 60·8%, P < 0·001).

Using a single ICD code to identify cases of HS had a high PPV (96·9%) but low sensitivity (63·4%) when including patients who lacked an HS ICD code but had mention of the diagnosis and sufficient information to clinically diagnose HS in their chart (Table 1). There was a statistically significant association between year of diagnosis and accuracy of case status (P = 0·03), although the PPV of having a single ICD code remained very high throughout the study period (93–99% per 5-year grouping). Most cases did not have HS-related CPT codes (1013 of 1316, 77·0%).

Table 1.

Sensitivity and positive predictive value of different cut points based on the number of International Classification of Diseases (ICD) codes for hidradenitis suppurativa (HS)

ICD count	0	1	2	≥ 3
Case	482	271	189	113
Not case	295	21	5	0
Positive predictive value		0·979	0·989	0·997
Sensitivity		0·634	0·428	0·284

In this large, single-institution study of HS, the presence of one HS ICD code had a high PPV for identifying cases of HS, but low sensitivity. More than one-third of cases had no HS-specific diagnostic codes. Our findings suggest that incidence studies relying solely on ICD codes will greatly undercount the true number of cases. Female patients, older patients, and those with a shorter duration of follow-up within our EHR were less likely to have an accurate HS ICD code. Although this was a single-centre study with relatively short follow-up, these findings align with the known problem of delayed diagnosis in HS. Developing methods to identify undiagnosed cases in the EHR could lead to earlier intervention and potential improvement of outcomes in HS.⁵

Data availability statement:

Per VUMC policy, individual-level data cannot be shared