Abstract The object of this study was to evaluate the sensi- tivity and positive predictive value (PPV) of International Classification of Diseases, 9th revision (ICD-9) codes 430–438 in the Sistema Informativo Sanitario Regionale (SISR), an Italian health care automated database. We com- pared the SISR with a manual search of all cases of transient ischaemic attack (TIA) and stroke discharged from the Novara Hospital, NW Italy. Results were as follows: SISR list: 1017 patients; manual list 1005. Linked: 896; false neg- atives: 109; false positives: 121. Sensitivity of codes 430–438: 77% at the primary position only and 89% at either the primary or secondary position; PPV: 93% and 88%. Sensitivity and PPV for specific codes vs. each subcategory (sensitivity at the primary position only/any position; PPV at the primary position only/any position): for 430, subarach- noid haemorrhage (33/35%; 46/43%); for 431, cerebral haemorrhage (57/59%; 77/75%); for 434, cerebral infarction (35/37%; 90/87%); for 436, stroke of unknown type (29/29%; 19/16%); and for 435, TIA (75/82%; 80/78%). The SISR database has a high PPV; sensitivity is high for TIA, but low for specific stroke ICD codes. Key words TIA • Stroke • ICD codes • Discharge • Accuracy • Sensitivity • Positive predictive value Introduction The use of administrative databases for evaluating the quali- ty of care for stroke patients is still uncommon [1–4]. The feasibility of studies relying on such databases mostly depends on the validity and reliability of the data recorded. Few studies have evaluated the validity of databases using the International Classification of Diseases, 9th revision (ICD-9) codes in stroke patients by comparing the informa- tion from recorded discharge diagnoses with medical charts. Sensitivity is generally high [4–9], but positive predictive value (PPV) depends on the inclusion of first-ever or recur- rent strokes, single codes evaluated or whether codes in the primary or secondary position are considered. We aimed to evaluate sensitivity and PPV of the ICD-9 codes for identi- fying cases of transient ischaemic attack (TIA) and stroke in the Sistema Informativo Sanitario Regionale (SISR), a health care automated database of the Piedmont Region, NW Italy. The SISR was created in 1976 and gathers several data including the international codes for diseases and proce- dures; the ICD-9 codes were replaced by ICD-9-CM in 2000. Methods We compared the SISR with a manual search of all cases of TIA and stroke discharged in the same year from the Novara Hospital, the second largest hospital of the Region. This is a 900-bed first- referral General Hospital with 36 000 admissions per year and a catchment area of about 300 000 persons. The SISR This database is maintained by the Piedmont Region’s Department of Health and includes all hospital admissions in the Region. It comprises information on the personal demographics of enrolees, admission and discharge date, vital status, hospital and Neurol Sci (2004) 25:281–288 DOI 10.1007/s10072-004-0355-8 M.A. Leone • A. Capponi • C. Varrasi • R. Tarletti • F. Monaco Accuracy of the ICD-9 codes for identifying TIA and stroke in an Italian automated database ORIGINAL Received: 14 June 2004 / Accepted in revised form: 12 November 2004 M.A. Leone () • C. Varrasi • R. Tarletti • F. Monaco Clinica Neurologica Ospedale Maggiore della Carità and Università del Piemonte Orientale “A. Avogadro” C.so Mazzini 18, I-28100 Novara, Italy e-mail: leomau@iol.it A. Capponi Ufficio Qualità Direzione Sanitaria Ospedale Maggiore della Carità Novara, Italy