We studied a large sample of young Indian adults who had serial measurements of BMI in childhood. As children, they were thin by international standards, but as adults had a high prevalence of overweight/obesity, diabetes and metabolic syndrome, an increasingly common scenario in urban developing country populations. Accelerated BMI gain during childhood was associated with an increased risk of adult IGT/DM and MS, probably mediated by increased adult adiposity7
. Our main objective was to determine whether screening tests could be derived from serial childhood BMI measurements to detect high-risk children. Data from the US Bogalusa Heart Study was used to predict risk of MS from single childhood BMI values15
but as far as we know, this is the first such analysis using longitudinal BMI data.
Strengths of the study were that it was population-based, and anthropometric data were collected by trained personnel at unusually frequent intervals. Only 19% of the original cohort participated, and they are likely to be unrepresentative. However, the differences in their mean childhood BMI from the rest of the cohort, though statistically significant, were small.
Predictions for MS were superior to those for IGT/DM. Sensitivity was 30-45% for MS and 29-37% for IGT/DM. Positive predictive values ranged from 32-41% for MS and 18-20% for IGT/DM. Thus a substantial proportion, though a minority, of ‘test-positive’ children developed disease, and a positive test could thus cause unwarranted anxiety for families. Specificity and negative predictive values were high (76-81% for MS and 72-87% for IGT/DM) indicating that a negative test is reassuring. The attributable fraction (9-24% for MS, 6-12% for IGT/DM) and likelihood ratio (1.4-2.0 for MS and 1.1-1.6 for IGT/DM) were low.
The longitudinal test gave superior predictions to those based on single overweight/obese BMI values, currently often used to identify children needing intervention. The latter, though highly specific (>97%), picked up few of the children at risk and thus had a low sensitivity (≤7%). This differs from findings in the Bogalusa study, in which a single overweight/obese BMI value recorded between the ages of 4 and 15 years had a sensitivity of 38% and specificity of 87% for predicting adult MS15
. If a lower BMI cut-off was used in Delhi, predictions were closer to those obtained using the combined test, but sensitivity remained lower.
Opinions will vary on the usefulness of these predictions. Likelihood ratios of >5.0 or <0.2 are generally considered important clinically16
and those for our combined test ranged from 1.1 to 2.0. No single predictive test can be taken in isolation, and likelihood ratios were similar for overweight/obesity, which most clinicians would consider worthy of intervention. Likelihood ratios do not change according to disease prevalence, and a 10 to 100% increase in the odds of developing disease is important if that disease is common and serious. Considering that these are predictions across an intervening period of several decades, we think they are impressive, although better predictive indices, possibly metabolic profiles in adolescence, should be sought.
Given the very high prevalence of disease in young adulthood in this population, it could be argued that screening tests are redundant; there should clearly be population-wide
attempts to reduce adiposity. Such interventions, however, have not been adopted even in rich countries, and are unlikely to happen quickly in developing countries. Parents in India who can afford to, do take their children to paediatricians regularly, and individual-level risk-assessment could be useful. Since observational studies cannot prove causation, intervening to limit BMI gain will not necessarily prevent adult disease. Furthermore, no consistently effective interventions have been identified17-20
. However, the seriousness of the diabetes epidemic in India demands the formulation of policy and clinical guidelines based on best evidence. Although randomised trials extending from childhood into adulthood have not been performed, lifestyle intervention trials have documented short-term improvements in adiposity, metabolic parameters and endothelial function17-22
. A commonsense approach would be to suggest that parents of test-positive children should be advised to consider lifestyle changes, to continue monitoring the child’s BMI, and to obtain relevant investigations (blood pressure, lipid profile and glucose concentrations).
Our analysis raises several practical issues. It would be possible to identify ‘test-positive’ children using a conventional BMI chart (crossing centiles upwards and above the 50th percentile at the later measurement). These changes could, however, appear subtle on a standard chart, and furthermore, clinicians may not be concerned about a child of average BMI, even if climbing centiles. Our new charts show upward and downward movement more clearly. They are more complex than standard charts, however, and clinicians would need additional explanation/guidance to use them. Since detailed growth data is not available for all populations, an external reference would be advantageous and also facilitate global comparisons. We adapted our test to overcome the large BMI differences between the Delhi and CDC populations. The CDC reference, however, produced widely varying predictions because these differences varied with age. Our results suggest that ‘local’ reference data, based on children as similar as possible to those in the population under consideration, are best for identifying high-risk children. We cannot, therefore, recommend the clinical use of our chart in non-Indian populations, or in children that differ greatly in BMI from the Delhi cohort. Comparable analyses need to be carried out, and charts derived, in different populations. Because our measurements were obtained as part of a research study, they are likely to have been more precise, resulting in better predictions, than could be achieved in clinical practice. However, in a simulation exercise, introducing random BMI measurement error, and misclassification of cases and controls, predictive parameters were little changed (details available on request).
Some of the predictive indices (positive and negative predictive values and attributable fraction) are influenced by the prevalence of the outcome. Although the prevalence of metabolic syndrome and IGT/DM was already high in the Delhi cohort, it will increase further with time. We are currently re-studying the cohort after an interval of 5 years, and will re-assess the predictions.