Databases and Diagnosis of Obesity: Pitfalls and Potential of Using ICD-9 Codes
Article Outline
Abbreviations: BMI, Body mass index, ICD, International Statistical Classification of Diseases and Related Health Problems
The effects of obesity are multiple and include not only medical, psychological, and economic consequences in obese individuals, but also healthcare and productivity costs to society. Because of the condition's high and rising prevalence, society is interested in healthcare utilization to measure both the health and economic impacts. Although in individual children, the most obvious obesity-related medical problems may emerge in later years, the investigation of groups of children is undertaken to gain insight into the current health burden. The study of hospital utilization has the potential to identify subtle differences not easily recognized in individual children. For instance, if obese children have a slightly increased risk of complications for a condition, then a small impact (eg, a 3-hour-longer average length of stay than in healthy-weight children) will go unrecognized unless groups of patients are examined together.
See related article, p 327
Administrative databases are potentially powerful tools because they capture discrete information, including clinical information, from many patients. The study by Woo et al1 demonstrates the potential pitfalls of this approach, however. Previous reports have linked childhood obesity based on International Statistical Classification of Diseases and Related Health Problems (ICD)-9 codes with utilization patterns,2 and Woo et al examine the difference between obesity measured from weight and height and obesity assigned by ICD-9 codes as these measures relate to hospital utilization.
The authors took advantage of 2 important characteristics of Cincinnati Children's hospital. First, the hospital routinely measures and enters into an electronic health record the weights and heights of admitted patients, thereby providing a classification of patient body mass index (BMI) category based on measurements rather than on a diagnosis code. Second, the hospital is the exclusive tertiary care provider for children in the area, which reduces the likelihood of variation from differing service patterns or patient variation that may occur when a hospital shares a “market” with other children's hospital services.
Most importantly, the study found that a diagnosis code of obesity was assigned to <10% of the children with a BMI ≥95th percentile. This finding is not surprising, given that rates of obesity diagnosis have been much lower than prevalence rates of obesity for many years.3, 4 But documenting this discrepancy is important and is now possible with electronic health records. Surprisingly, those obese children with an obesity ICD-9 code had an overall lower utilization than nonobese children, as measured by per-person length of stay and per-person average number of discharges; however, children with BMI ≥95th percentile but with no obesity ICD-9 code had significantly longer per-person length of stay but slightly fewer per-person average number of discharges than nonobese children. The authors also found unexpected relationships between each of 13 primary diagnosis categories and diagnosed or undiagnosed obesity. Although the analysis revealed higher odds of an endocrine/metabolic category diagnosis among those discharges with diagnosed obesity, these odds were significantly lower in those with undiagnosed obesity. The diagnosed obesity discharges had higher odds of a primary diagnosis of a mental health condition.
A limitation of this study is the substantial proportion (30%) of records of clinical stays of patients age 2 to 20 years in which weight or height was missing. Missing weight or height could be systematically related to the severity of illness or type of primary diagnosis and thus utilization. In addition, about 30% of the non–neonatal intensive care unit beds in this hospital are for psychiatric and rehabilitation care. Utilization patterns for these units may differ from the general medicine and surgery units, but unit utilization patterns were not reported. As a result, the overall characteristics of discharges by obesity category may not generalize to tertiary care children's hospitals with different allocations of beds.
This study proves what was expected: using ICD-9 codes is a poor way to track obesity prevalence in hospitalized children. More importantly, the codes will not generate a patient sample that reflects the larger group of patients with measured obesity, as demonstrated by the inconsistent relationship between utilization and either diagnosed obesity or undiagnosed obesity.
The relationships among utilization, primary diagnosis category, obesity, and obesity diagnosis are complex and subject to human bias. For this reason, inferences of clinical meaning from the diagnostic codes created primarily for administrative purposes are misleading. The practitioner's perception of the relationship between the primary diagnosis and the child's weight status will influence the decision of whether or not to add obesity as a second or third diagnosis. A provider may be disposed toward seeing a potential connection between weight status and mental health, explaining in part the higher odds of a mental health primary diagnosis in children with ICD-9 obesity diagnosis. It would be useful to examine the likelihood of a primary diagnosis of mental health among all obese patients (diagnosed or not) compared with a nonobese reference group. Similarly, in the group of obese children with hospital stays for a primary diagnosis of an endocrine/metabolic disorder (eg, type 2 diabetes), increased awareness of a connection between obesity and the disorder may lead the practitioner to actively diagnose obesity, leaving a smaller pool of obese patients with a metabolic/endocrine disorder who have no obesity diagnosis. Statistically, these obese patients without an ICD-9 obesity diagnosis have lower odds of metabolic/endocrine disorder diagnosis, but in reality this relationship may instead reflect the higher likelihood of patients with metabolic/endocrine disorders to receive an obesity diagnosis.
Regardless of whether or not obesity is a coded diagnosis, one might expect overall higher utilization in obese patients if obesity is associated with worse medical state. Although this study found fewer inpatient discharges in obese patients, we cannot conclude that these patients are healthier. When a medical condition commonly associated with obesity is addressed primarily in outpatient settings (eg, type 2 diabetes), the lower hospital utilization in obese patients may reflect patterns of utilization for common conditions linked to obesity relative to utilization for other conditions not related to obesity. Bundling a wide variety of conditions within broad primary diagnosis categories, such as respiratory diagnosis, further obscures any relationship between obesity and health state; for example, obese patients may have higher utilization for asthma than nonobese patients, but because obese patients do not have cystic fibrosis, the possible positive relationship between obesity and asthma hospital stays will be cancelled out by the inverse association between obesity and cystic fibrosis stays. Tying measured obesity to specific diagnoses and also to more comprehensive utilization patterns, including outpatient utilization as well as costs or charges, will more accurately reflect the health state and cost of obesity.
The authors' main conclusions are sound; identification of obesity based on measured weight and height demonstrates higher prevalence and different utilization patterns than identification of obesity from ICD-9 codes. The better validity of measured obesity is obvious. But this measure could be used more effectively in 2 ways: first, by documenting BMI in all hospitalized patients, which would extend the recent inclusion of BMI as a Healthcare Effectiveness Data and Information Set (HEDIS) measure in outpatient care, and second, by grading severity within the large group of children with BMI >95th percentile. To tap the potential of measured obesity, the variables related to BMI status also need to be more precise than the broad diagnosis categories used here. Electronic health records hold promise for better database research on disease and utilization because their purpose—health assessment and care—is more closely aligned to questions of obesity and health status than administrative databases. But they remain subject to human bias, and thus findings must be interpreted cautiously.
References
- . Obesity identified by discharge ICD-9 codes underestimates the true incidence of obesity in hospitalized children. J Pediatr. 2009;154:327–331
- . Economic burden of obesity in youths aged 6 to 17 years: 1979-1999. Pediatrics. 2002;109:e81
- . Screening and counseling associated with obesity diagnosis in a national survey of ambulatory pediatric visits. Pediatrics. 2005;116:112–116
- . Resource utilization and expenditures for overweight and obese children. Arch Pediatr Adolesc Med. 2007;161:11–14
PII: S0022-3476(08)01077-9
doi:10.1016/j.jpeds.2008.12.007
© 2009 Mosby, Inc. All rights reserved.
Refers to article:
- Obesity Identified by Discharge ICD-9 Codes Underestimates the True Prevalence of Obesity in Hospitalized Children , 28 October 2008
