This study examined for the first time how well an ICD code of obesity identifies obese children in Canadian administrative health data. The sensitivity of an ICD code-based obesity diagnosis for detecting measured obesity was low, only 7% of obese children were correctly identified. Those correctly identified had higher health care utilization than those without an ICD diagnosis of obesity.
An ICD diagnosis of obesity has been used previously to identify obese children in administrative databases [6–10]. A recent study showed that the majority of children (> 90%) with measured obesity did not receive an ICD diagnosis of obesity during their inpatient stay at a tertiary care hospital in Ohio . We had speculated that the sensitivity of an ICD diagnosis might be higher if both physician visits and hospital stays are used since children see their general practitioner or pediatrician far more often than being admitted to a hospital, and a weight problem may be more likely to be picked up during a well child visit or a consultation for a minor ailment. This hypothesis is supported by the fact that all children in our sample received the ICD diagnosis of obesity during a physician visit (either general practicioner or pediatrician).
The consequence of the poor sensitivity of the ICD code for obesity is that administrative data will grossly underestimate the true population prevalence of obesity. We were also interested in examining whether this misclassification is differential with regard to health care utilization. When comparing the costs in obese children and normal weight children, costs were 16% higher in children with measured obesity. This cost differential increased to 108% when the analysis was based on an ICD diagnosis of obesity between 2002 and 2004. That is, using an ICD diagnosis compare health care utilization between obese and non-obese children severely overestimated health care costs for obese children. A possible explanation is that physicians are more likely to diagnose obesity in a child if multiple, potentially obesity-related, co-morbidities are present. This argument is further supported by the finding that the number of physician visits was significantly higher in those correctly identified as obese. An alternative explanation could be that with the identification of obesity, physicians scheduled more frequent evaluations of the patients. However, a comparison of the number of physician visits three years before and after the ICD diagnosis showed no difference (p = 0.91, data not shown). Hampl et al.  examining inpatient utilization in a pediatric primary care centre in the US reported higher health care costs for children with diagnosed obesity compared to those with undiagnosed obesity. By contrast, Woo et al.  found that children with diagnosed obesity had shorter hospital stays and fewer hospital discharges than both non-obese and undiagnosed obese patients. The apparent discrepancy may be explained by the fact that the study was done in a tertiary care hospital while some health conditions that are more common in obese children are primarily treated on an outpatient basis (e.g. asthma, type 2 diabetes). Children with diagnosed obesity in Woo's study were more likely to have primary diagnoses of mental health, endocrine, and musculo-skeletal disorders compared to children with undiagnosed obesity. This finding may indicate that the presence of a 'typical' obesity-related disorder increases the likelihood of receiving a diagnosis of obesity .
Besides the methodological aspects, our findings raise some concern about identification of obesity in the primary care system. General practitioners and pediatricians have a critical role in the diagnosis, education and management of overweight and obesity as they constitute the first point of contact within the health care system. An obese child that is not diagnosed (and not counseled) is a lost opportunity for secondary prevention. As shown in our analysis, less than 10% of obese children are diagnosed as obese by an ICD code. Even more concerning is that a quarter of children with undiagnosed obesity had a BMI between 29 and 44 kg/m2, which is well beyond the age-specific obesity cut-off of approximately 23 kg/m2  and clearly associated with health risks. The marked discrepancy between the ICD-based prevalence between 2002 and 2004 (1.5%) and that of measured obesity (16.4%) suggests that obesity was frequently overlooked or the issue was avoided by the physicians. On the other hand, ICD codes in administrative health data are primarily collected for billing purposes and the lack of an ICD code of obesity for an obese child may not necessarily indicate that a physician did not address the problem in the consultation. One may also argue that the Canadian health care system does not have the capacity to manage childhood overweight and obesity, and that the problem is best tackled by primary prevention measures.
The strengths of the current paper are the use of longitudinal administrative data from a universal single provider health care system linked with a population-based survey, and the coverage of both physician visits and hospital stays. Our findings are limited by the single BMI measurement at age 10/11 years and the lack of synchronicity between the BMI measurement and the physician visit/hospital stay. However, obesity is the result of long-term lifestyle habits and not expected to change within a relatively short time frame. Another limitation is the response rate of 51%, which may have resulted in a selected sample. If the non-responders in the survey had a different probability of being identified as obese by their physicians compared to the children participating in the survey, the results would be biased. On the other hand, sensitivity and specificity are not affected by the prevalence of a condition, and therefore a higher or lower prevalence of obesity in the non-responders would not influence the results. The non-responders would also be likely to be seen by the same physicians as their participating peers and hence the sensitivity and specificity can be expected to be comparable between the two groups.