Neurology -- Correspondence for Golomb et al., 67 (11) 2053-2055

Correspondence published:

Accuracy and yield of ICD-9 codes for identifying children with ischemic stroke: Daniel L. Labovitz (31 January 2007)

Reply from the Authors: Meredith R. Golomb, Bhuwan P. Garg, Chandan Saha, and Linda S. Williams (31 January 2007)

Accuracy and yield of ICD-9 codes for identifying children with ischemic stroke 31 January 2007

Daniel L. Labovitz,
NYU School of Medicine
Stroke Center, HCC 5F, 530 First Ave, New York, NY 10016
Send Correspondence to journal:
Re: Accuracy and yield of ICD-9 codes for identifying children with ischemic stroke

daniel.labovitz{at}med.nyu.edu Daniel L. Labovitz

Golomb et al address an important issue in stroke epidemiology: the accuracy of stroke ICD9 codes which appears to be even worse in children than adults.[1]
As hospitals move to electronic medical records, allowing electronic access to more clinical detail, we will be ever more tempted to use large administrative datasets for clinical research. Such datasets may be especially valuable for investigation of rare diseases such as childhood stroke. Unfortunately, the stroke ICD9 codes are crude and often are applied erroneously, making the systematic capture of true stroke cases difficult. Despite a valiant effort, it appears that this study accidentally may have revealed some of the reasons why it is so difficult.
The authors use the word “accuracy” in place of the correct label "positive predictive value", which in turn is just one of several possible tests for accuracy. Using the wrong words makes a difficult topic even more difficult.
Were the authors seeking only acute stroke cases? Their definition of stroke only as “radiographic evidence of infarction” and their search for stroke cases diagnosed with cerebral palsy suggest they mixed in chronic cases as well. Hospital ICD9 codes will perform poorly at identifying prevalent stroke cases, especially if the definition of stroke requires brain imaging.
In October 1992, stroke ICD9 codes were modified to use 5 digits but the paper refers only to 3-digit codes. For 433.xx and 434.xx it is the fifth digit which indicates whether it is an ischemic stroke case or merely arterial stenosis or occlusion without infarction. Did the authors include all patients without reference to the 5th digit? Such an approach will by definition include non-stroke cases, making the positive predictive value unnecessarily low.
The authors did not cite an excellent paper which evaluated the utility of ICD9 stroke codes in a large, population-based stroke incidence survey.[2] That study was notable for finding that adding secondary ICD9 discharge codes only marginally increased sensitivity and that combining surveillance of outpatient emergency department visits with hospital ICD9 discharge codes identified fully 89% of hospitalized and non-hospitalized stroke cases.
References
1. Golomb MR, Garg BP, Saha C, Williams LS. Accuracy and yield of ICD-9 codes for identifying children with ischemic stroke. Neurology 2006;67:2053-2055.
2. Piriyawat P, Šmajsová M, Smith MA, et al. Comparison of active and passive surveillance for cerebrovascular disease: the Brain Attack Surveillance in Corpus Christi (BASIC) Project. Am J Epidemiol 2002;156:1062–1069.
Disclosure: The author reports no conflicts of interest.

Reply from the Authors 31 January 2007

Meredith R. Golomb,
Indiana University School of Medicine, Department of Neurology, Division of Pediatric Neurology
575 West Dr. Building XE-040, Indianapolis, IN 46202,
Bhuwan P. Garg, Chandan Saha, and Linda S. Williams
Send Correspondence to journal:
Re: Reply from the Authors

mgolomb{at}iupui.edu Meredith R. Golomb, et al.

We thank Dr. Labovitz for his comments on our study [1] and for bringing the paper by Piriyawat et al to our attention. [2]
We used the term “accuracy” rather than “positive predictive value” because “positive predictive value” depends on sensitivity, specificity and prevalence of a disease. The prevalence of pediatric arterial ischemic stroke is not clear; estimates of the incidence of pediatric stroke in previous studies have ranged from 2.1 per 100,000 children per year to 13 per 100,000 children per year. [3-6]
In our study, subject selection was ICD-9 code-based and would provide a biased estimate of prevalence. If we used our sample estimate of prevalence to calculate positive predictive value, this would result in a biased estimate of positive predictive value. Our objective was to demonstrate that ICD-9 codes for pediatric stroke are often incorrectly assigned.
We looked at children with stroke at any time. If we had looked only at acute stroke cases, the ICD-9 codes would have performed even more poorly. We did not look at the fifth digit in the codes.
References
3. Chung B, Wong V. Pediatric stroke among Hong Kong Chinese subjects. Pediatrics 2004;114:e206-212.
4. Fullerton HJ, Wu YW, Zhao S, Johnston SC. Risk of stroke in children: Ethnic and gender disparities. Neurology 2003;61:189-94.
5. Broderick J, Talbot GT, Prenger E, Leach A, Brott T. Stroke in children within a major metropolitan area: the surprising importance of intracerebral hemorrhage. Journal of Child Neurology 1993;8:250-255.
6. Giroud M, Lemesle M, Gouyon JB, Nivelon JL, Milan C, Dumas R. Cerebrovascular disease in children under 16 years of age in the city of Dijon, France: a study of incidence and clinical features from 1985 to 1993. J Clin Epidemiol 1995;48:1343-1348.
Disclosure: The authors report no conflicts of interest.