Abstract
Objectives: Hospital Episode Statistics (HES) are routinely recorded at every hospital admission within the National Health Service (NHS) in England. This study validates diagnostic ICD-10 codes within HES as a method of identifying cases of idiopathic inflammatory myopathies (IIMs).
Methods: All in-patient admissions at one NHS trust between 2010-2020 with relevant diagnostic ICD10 codes were extracted from HES. Hospital databases were used to identify all outpatients with IIM and electronic care records reviewed to confirm coding accuracy. Total hospital admissions were calculated from NHS Digital reports. Sensitivity and specificity of each code and code combinations
were calculated to develop an optimal algorithm. The optimal algorithm was tested in a sample of admissions at another NHS Trust.
Results: 510/672 individuals identified by HES were confirmed to have IIM. Overall, positive predictive value (PPV) is 76% and sensitivity is 89%. Combination algorithms achieved PPVs between 89-94%. HES can also predict the presence of IIM-associated ILD to a 79% PPV and 71% sensitivity. The optimal algorithm excluded children (except JDM code M33.0) and combined M33.0, M33.1,
M33.9, M36.0, G72.4, M60.8 and M33.2, and included M60.9 only if it occurs alongside an ILD code (J84.1, J84.9 or J99.1). This produced an 88.9% PPV and 84.2% sensitivity. Retesting this algorithm at another NHS Trust confirmed a high PPV (94.4%).
Conclusion: IIM ICD-10 code combinations in HES have high positive predictive values, and sensitivities. Algorithms tested in this study could be applied across all NHS trusts to enable robust and cost-effective whole-population research into the epidemiology of IIM.
Methods: All in-patient admissions at one NHS trust between 2010-2020 with relevant diagnostic ICD10 codes were extracted from HES. Hospital databases were used to identify all outpatients with IIM and electronic care records reviewed to confirm coding accuracy. Total hospital admissions were calculated from NHS Digital reports. Sensitivity and specificity of each code and code combinations
were calculated to develop an optimal algorithm. The optimal algorithm was tested in a sample of admissions at another NHS Trust.
Results: 510/672 individuals identified by HES were confirmed to have IIM. Overall, positive predictive value (PPV) is 76% and sensitivity is 89%. Combination algorithms achieved PPVs between 89-94%. HES can also predict the presence of IIM-associated ILD to a 79% PPV and 71% sensitivity. The optimal algorithm excluded children (except JDM code M33.0) and combined M33.0, M33.1,
M33.9, M36.0, G72.4, M60.8 and M33.2, and included M60.9 only if it occurs alongside an ILD code (J84.1, J84.9 or J99.1). This produced an 88.9% PPV and 84.2% sensitivity. Retesting this algorithm at another NHS Trust confirmed a high PPV (94.4%).
Conclusion: IIM ICD-10 code combinations in HES have high positive predictive values, and sensitivities. Algorithms tested in this study could be applied across all NHS trusts to enable robust and cost-effective whole-population research into the epidemiology of IIM.
| Original language | English |
|---|---|
| Journal | Rheumatology Advances in Practice |
| Publication status | Accepted/In press - 25 Oct 2022 |