A study to determine the sensitivity and specificity of hospital discharge diagnosis data used in the MICA studyстатья из журнала
Аннотация: To determine the sensitivity and specificity of each ICD9 code for a diagnosis of definite or possible myocardial infarction (MI) from the perspective of the Myocardial Infarction Causality Study (MICA) and to use these data to estimate the likely number of MICA cases in Scotland that would be undetected were these codes omitted from the study.Women resident and registered with general practitioners in the Tayside region of Scotland between October 1993 and October 1995.All SMR1 records of Tayside hospitalizations containing ICD9 (International Classification of Diseases, ninth revision) codes for myocardial infarction (410) or possible myocardial infarction (411, 412, 413, 414, 427.4, 427.5, 786.5) were identified for women aged between 16 and 44 years between 1 October 1993 and 15 October 1995. Original case records were sought and each episode abstracted using a predefined form. Records were independently scrutinized by two consultant cardiologists blinded to the SMR1 code. Cases were categorized as definite MI, possible MI or unlikely MI. Where there was disagreement between the two cardiologists, the profiles for such events were examined by a third cardiologist who acted as the final adjudicator. The adjudicator's verdict was, in this study, considered dominant. The sensitivity, specificity and positive predictive value of each ICD9 code was determined.Two hundred and fifty-three women fulfilled the SMR1 search criteria. Case records of 204 (81%) were retrieved but four case records contained no data on the admission of interest and were classified as invalid. Forty-six of the 200 remaining patients were ineligible for the MICA study leaving 154 records for evaluation. There were 12 patients who had a discharge code for MI (ICD9 410). Of these, 11 were judged as a definite MI by both cardiologists. One event (discharge code ICD9 410) was judged as 'possible' by one cardiologist and 'unlikely' by the other. The adjudicator subsequently judged this event as 'definite'. Another six events were subsequently judged as 'possible'. Thus, after adjudication, 12 cases of definite MI and six cases of 'possible' MI were identified. The sensitivity and specificity of ICD9 code 410 was 67% and 100% respectively. The positive predictive value was 100%. The sensitivity of code 411 was 5.6%. The specificity was 99% and the positive predictive value was 50%. Code 413 had a sensitivity of 5.6% with a specificity of 94% and a positive predictive value of 9.1%. Code 414 also had a sensitivity of 5.6%. The specificity was 86% and the positive predictive value was 4.5%. Code 786.5 had a sensitivity of 17%, a specificity of 23% and a positive predictive value of 2.5%. Code 427.5 failed to identify any definite or possible cases.In the MICA Study, ICD9 code 410 was found to be the most robust. All 12 patients judged to have had a definite MI had the appropriate discharge code (ICD9 410). The six patients judged to have had a possible MI all had discharge codes other than that for MI (410). However, identifying these six patients required the validation of a further 160 events-giving a combined sensitivity of 33%, a specificity of 0% and a positive predictive value of only 3.8%. The use of ICD9 codes 411, 413, 414, 427.5 and 786.5 must, therefore, only be employed when circumstances fully justify the additional workload.
Год издания: 1998
Издательство: Wiley
Источник: Pharmacoepidemiology and Drug Safety
Ключевые слова: Machine Learning in Healthcare, Emergency and Acute Care Studies, Nursing Diagnosis and Documentation
Другие ссылки: Pharmacoepidemiology and Drug Safety (HTML)
PubMed (HTML)
PubMed (HTML)
Открытый доступ: closed
Том: 7
Выпуск: 5
Страницы: 311–318