Accuracy of ICD-9-CM codes for identifying cardiovascular and stroke risk factors

Washington University in St. Louis, San Luis, Missouri, United States
Medical Care (Impact Factor: 3.23). 05/2005; 43(5):480-5.
Source: PubMed


We sought to determine which ICD-9-CM codes in Medicare Part A data identify cardiovascular and stroke risk factors.
This was a cross-sectional study comparing ICD-9-CM data to structured medical record review from 23,657 Medicare beneficiaries aged 20 to 105 years who had atrial fibrillation.
Quality improvement organizations used standardized abstraction instruments to determine the presence of 9 cardiovascular and stroke risk factors. Using the chart abstractions as the gold standard, we assessed the accuracy of ICD-9-CM codes to identify these risk factors.
ICD-9-CM codes for all risk factors had high specificity (>0.95) and low sensitivity (< or =0.76). The positive predictive values were greater than 0.95 for 5 common, chronic risk factors-coronary artery disease, stroke/transient ischemic attack, heart failure, diabetes, and hypertension. The sixth common risk factor, valvular heart disease, had a positive predictive value of 0.93. For all 6 common risk factors, negative predictive values ranged from 0.52 to 0.91. The rare risk factors-arterial peripheral embolus, intracranial hemorrhage, and deep venous thrombosis-had high negative predictive value (> or =0.98) but moderate positive predictive values (range, 0.54-0.77) in this population.
Using ICD-9-CM codes alone, heart failure, coronary artery disease, diabetes, hypertension, and stroke can be ruled in but not necessarily ruled out. Where feasible, review of additional data (eg, physician notes or imaging studies) should be used to confirm the diagnosis of valvular disease, arterial peripheral embolus, intracranial hemorrhage, and deep venous thrombosis.

  • Source
    • "HF and CM events and comorbidities were ascertained through administrative codes and not confirmed clinically. However, administrative codes for HF and cardiovascular comorbidities have high specificity (≈95%) and positive predictive value (95%).(2005)–(2012) In addition, measures of left ventricular systolic function were not available, and neither the severity of CM nor differentiation of systolic from diastolic HF could be established. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Adjuvant trastuzumab improves survival for women with human epidermal growth factor receptor 2-positive breast cancer, but increases risk for heart failure (HF) and cardiomyopathy (CM). However, clinical trials may underestimate HF/CM risk because they enroll younger subjects with fewer cardiac risk factors. We sought to develop a clinical risk score that identifies older women with breast cancer who are at higher risk of HF or CM after trastuzumab. Using the Surveillance, Epidemiology and End Results (SEER)-Medicare database, we identified women with breast cancer who received adjuvant trastuzumab. Using a split-sample design, we used a proportional hazards model to identify candidate predictors of HF/CM in a derivation cohort. A risk score was constructed using regression coefficients, and HF/CM rates were calculated in the validation cohort. The sample consisted of 1664 older women (mean age 73.6 years) with 3-year HF/CM rate of 19.1%. A risk score consisting of age, adjuvant chemotherapy, coronary artery disease, atrial fibrillation or flutter, diabetes mellitus, hypertension, and renal failure was able to classify HF/CM risk into low (0 to 3 points), medium (4 to 5 points), and high (≥6 points) risk strata with 3-year rates of 16.2%, 26.0%, and 39.5%, respectively. A 7-factor risk score was able to stratify 3-year risk of HF/CM after trastuzumab between the lowest and highest risk groups by more than 2-fold in a Medicare population. These findings will inform future research aimed at further developing a clinical risk score for HF/CM for breast cancer patients of all ages.
    Full-text · Article · Dec 2014 · Journal of the American Heart Association
  • Source
    • "Trained medical coders review the information in the patient record for a clinical episode and assign a set of appropriate ICD9 codes. Manual coding can be noisy: human coders sometimes disagree,9 tend to be more specific than sensitive in their assignments,10 and sometimes make mistakes.11 12 Nevertheless, the large set of narratives and their associated ICD9 codes, especially when taken in aggregate, represents a valuable dataset to learn from.13 "
    [Show abstract] [Hide abstract]
    ABSTRACT: The volume of healthcare data is growing rapidly with the adoption of health information technology. We focus on automated ICD9 code assignment from discharge summary content and methods for evaluating such assignments. We study ICD9 diagnosis codes and discharge summaries from the publicly available Multiparameter Intelligent Monitoring in Intensive Care II (MIMIC II) repository. We experiment with two coding approaches: one that treats each ICD9 code independently of each other (flat classifier), and one that leverages the hierarchical nature of ICD9 codes into its modeling (hierarchy-based classifier). We propose novel evaluation metrics, which reflect the distances among gold-standard and predicted codes and their locations in the ICD9 tree. Experimental setup, code for modeling, and evaluation scripts are made available to the research community. The hierarchy-based classifier outperforms the flat classifier with F-measures of 39.5% and 27.6%, respectively, when trained on 20 533 documents and tested on 2282 documents. While recall is improved at the expense of precision, our novel evaluation metrics show a more refined assessment: for instance, the hierarchy-based classifier identifies the correct sub-tree of gold-standard codes more often than the flat classifier. Error analysis reveals that gold-standard codes are not perfect, and as such the recall and precision are likely underestimated. Hierarchy-based classification yields better ICD9 coding than flat classification for MIMIC patients. Automated ICD9 coding is an example of a task for which data and tools can be shared and for which the research community can work together to build on shared models and advance the state of the art.
    Full-text · Article · Dec 2013 · Journal of the American Medical Informatics Association
  • Source
    • "By ignoring the unstructured text, we could be missing a substantial portion of adverse events17. Many studies18 have shown that coded information like ICD-9 are inadequate to accurately build patient cohorts and there is a considerable advantage19 in using the unstructured clinical text of EHRs. We argue that such an advantage would also extend to drug safety signal detection. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Drug-drug interactions (DDI) account for 30% of all adverse drug reactions, which are the fourth leading cause of death in the US. Current methods for post marketing surveillance primarily use spontaneous reporting systems for learning DDI signals and validate their signals using the structured portions of Electronic Health Records (EHRs). We demonstrate a fast, annotation-based approach, which uses standard odds ratios for identifying signals of DDIs from the textual portion of EHRs directly and which, to our knowledge, is the first effort of its kind. We developed a gold standard of 1,120 DDIs spanning 14 adverse events and 1,164 drugs. Our evaluations on this gold standard using millions of clinical notes from the Stanford Hospital confirm that identifying DDI signals from clinical text is feasible (AUROC=81.5%). We conclude that the text in EHRs contain valuable information for learning DDI signals and has enormous utility in drug surveillance and clinical decision support.
    Full-text · Article · Mar 2013
Show more