Related Experiment Video
Updated: Jul 10, 2026
Inverse Probability of Treatment Weighting (Propensity Score) using the Military Health System Data Repository and National Death Index
Published on: January 8, 2020
Validation of ICD-9 codes with a high positive predictive value for incident strokes resulting in hospitalization
Christianne L Roumie1, Edward Mitchel, Patricia S Gideon
1Veterans Administration, Tennessee Valley Healthcare System, Tennessee Valley Geriatric Research Education Clinical Center (GRECC), Nashville, TN 37212, USA. christianne.roumie@vanderbilt.edu
Insights
This study validated International Classification of Diseases, 9th Revision (ICD-9) codes for identifying incident strokes. An 80% positive predictive value was achieved by using primary discharge diagnoses and excluding prior outpatient stroke records.
Area of Science:
- Epidemiology
- Medical Informatics
- Neurology
Background:
- Accurate identification of incident strokes is crucial for epidemiological research and understanding disease etiology.
- International Classification of Diseases, 9th Revision (ICD-9) codes are widely used for disease surveillance and research, but their accuracy for specific conditions requires validation.
Purpose of the Study:
- To validate International Classification of Diseases, 9th Revision (ICD-9) codes for identifying incident strokes among Tennessee Medicaid enrollees.
- To determine the positive predictive value (PPV) of specific ICD-9 codes for incident stroke cases.
Main Methods:
- A cohort of Tennessee Medicaid enrollees aged 50-84 years hospitalized with stroke between 1999-2003 was identified using specific ICD-9 codes.
- Medical records of a systematic sample of 250 patients were reviewed to confirm incident stroke diagnoses.
- An algorithm was developed to refine the identification of incident strokes by considering primary discharge diagnoses and excluding prior outpatient stroke history.
Main Results:
- Of 231 reviewed charts, 205 confirmed new outpatient strokes (89%).
- Using primary discharge diagnosis alone yielded a 96% PPV, but this decreased to 74% when excluding recurrent strokes.
- An algorithm combining primary diagnosis and exclusion of prior outpatient diagnoses achieved an 80% PPV for incident strokes.
Conclusions:
- A validated strategy using primary discharge diagnosis and excluding prior outpatient diagnoses achieved an 80% PPV for incident strokes.
- This method provides a reliable cohort for etiologic studies, particularly investigating medication exposures, despite potential underestimation of incident stroke cases.
Purpose:
To validate ICD 9 codes with a high positive predictive value (PPV) for incident strokes. The study population consisted of Tennessee Medicaid enrollees aged from 50 to 84 years.
Methods:
We identified all patients who were hospitalized with a discharge diagnosis of stroke between 1999 and 2003 using highly specific codes (ischemic stroke ICD 9-CM codes 433.x1, 434 [excluding 434.x0], or 436; intracerebral hemorrhage [431]; and subarachnoid hemorrhage [430]). We reviewed medical records of a systematic sample of 250 cohort members. We randomly selected 10-30 eligible records for review from hospitals with at least 10 stroke hospitalizations.
Results:
We reviewed 231 charts (93% of total sampled), and 205 (89%) met study criteria for new outpatient stroke. Of the 205 confirmed new outpatient strokes, 196 had stroke listed as the primary discharge diagnosis (PPV = 96%). However, 46 (23%) of the 196 patients identified by the primary diagnosis also had a remote stroke history (recurrent stroke not incident). Thus the PPV of the primary discharge diagnosis for identifying incident stroke decreased to 74%. When we applied an algorithm that restricted our population to those with stroke as the primary diagnosis and excluded patients with any prior outpatient diagnosis of stroke, we identified incident stroke events with more precision (PPV = 80%).
Conclusion:
The PPV of incident strokes was 80% using our strategy of primary discharge diagnosis and excluding prior outpatient diagnoses of stroke. Although an unknown percentage of incident strokes are missed, this group of proven incident stroke patients can be used for etiologic studies of medication exposures.
