Methods to derive a modified mayo score from electronic health records data

Abstract BACKGROUND The modified Mayo Score (MMS), which consists of stool frequency, rectal bleeding, and endoscopic subscores, is the most commonly used method for measuring disease activity in Ulcerative Colitis (UC) patients. It is a standard outcome measure in clinical trials, and is recommended as a clinical trial endpoint by the US Food and Drug Administration. Each component has a score from 0-3; the total MMS ranges from 0 to 9. MMS is infrequently recorded in routine care, in part because it requires an endoscopic assessment. Real world data, including electronic health record (EHR) databases, are useful for studying longitudinal disease and treatment outcomes, but most EHR databases do not include Mayo scores. Building off the methods described by Rudrapatna et al., 2021, we calculated a derived MMS using data elements available in EHRs. METHODS Data were derived from the OM1 PremiOM UC Dataset (OM1, Boston, MA), a multisource real-world database with linked healthcare claims and EHR data on US patients with UC (2013-present). Each of the three subscores was extracted from clinical notes and endoscopy reports, either by automated natural language processing methods or by abstraction by a trained medical abstractor. Rectal bleeding and stool frequency scores occurring within 4 months before or up to 1 month after the endoscopic score were combined to calculate a MMS. Multiple MMSs could be calculated for each patient, but each component score could contribute to only one MMS. RESULTS Among 12,849 UC patients, only 38 had a full Mayo score recorded in their notes, and 1 had a modified Mayo recorded. After applying the above algorithm, we were able to calculate a derived MMS for 739 patients. 220 patients had > 1 score available. Figure 1 shows the distribution of derived MMSs. Patients with a MMS were younger and more likely to be Asian, more likely to be treated with biologic therapies and less likely to be treated with mesalamine compared with those who could not have a MMS calculated (Table 1). CONCLUSIONS A derived MMS can be calculated from data that may be recorded in the medical record. Further work to validate these measures against a gold standard should be considered, but was not feasible in the OM1 dataset due to the paucity of observed Mayo scores in EHR data. Increased emphasis on standardized collection of relevant information in EHR systems would further improve the availability of derived MMSs. The population for which a MMS can be derived may have more severe disease and more frequent interactions with the healthcare system as evidenced by higher proportion of patients prescribed biologic therapies. MMSs can be derived and used to describe UC patients and could potentially be used to study longitudinal changes in patient disease status and treatment effectiveness in settings where observed scores are not available.