Identification of individuals at risk for diabetes and CHD is the first step for primary prevention. Metabolic syndrome has received great fanfare for its putative value to identify at-risk patients for prevention interventions, despite a paucity of data about its actual performance in usual clinical care settings. We have demonstrated here that it is possible to identify patients at risk of developing diabetes and CHD by identifying risk factor clustering using a looser-than-formal metabolic syndrome definition based on combining formal and surrogate criteria available in the EHR of a large primary care network. A looser set of definitions was needed to account for the missing information that is characteristic of usual care data, especially obesity measures and indication of fasting status. Despite missing data, the validation study demonstrated that our approach to define EHR metabolic syndrome was 91% specific to identify patients with formally-diagnosed metabolic syndrome. Even a less specific, more sensitive approach to the definition identified patients at risk for adverse consequences of risk factor clustering.
Using a simple diagnostic algorithm, the patients identified in the EHR as having Metabolic syndrome
were more than 6 times more likely to develop diabetes, about 42% more likely to develop CHD, and to have higher health resource utilization and health care costs over three years of follow-up compared with individuals identified as not meeting metabolic syndrome diagnosis. The higher health care utilization costs in patients with metabolic syndrome are in accordance with other studies. Curtis et al
. found that individuals with the metabolic syndrome increased Medicare health care total costs by 20% to 30% [14
]. In individuals with all 5 criteria measured, ORs for increased risk of diabetes (OR = 8.67) and CHD (OR = 1.91) are also concordant with published literature: in studies with complete, standardized phenotyping individuals with the metabolic syndrome are about 3-6 times more likely to develop diabetes and to have twice the risk for CHD [1
]. As shown in other reports, our results argue in favor of metabolic syndrome being a stronger predictor of incident diabetes than CHD [16
]. In addition, we found that patients diagnosed with diabetes or CHD at baseline had about twice the utilization rates and costs compared with patients with EHR metabolic syndrome. These data support the notion that risk factor clustering is identifiable in usual clinical care, is associated with more adverse health outcomes over time, but is less costly than its full-blown diabetes and CHD outcomes. The data argue for the value of risk factor clustering as embodied in metabolic syndrome as a high-risk state amenable to and worthy of detection to prevent transitions from the lower-cost 'risk state' to the higher-cost 'outcome state' of chronic disease management.
Missing data could potentially have biased our findings. Using the data from patients having all five criteria measured and including patients with diabetes and CHD, the prevalence of EHR metabolic syndrome (23%) in our population was very similar to national data: in the NHANES, 20% to 25% of the US population had metabolic syndrome [3
]. Once we removed diabetes and CHD, the results from this subsidiary analysis (with all five criteria measured) was very similar to the results using our primary approach (with ≥ three criteria measured). Our approach allowing up to two missing characteristics makes greater use of the available clinical care data with no apparent cost to the validity of the approach. Indeed, our primary algorithm allowed us to classify three-quarters of all patients in the population into one of three metabolic syndrome categories despite the relatively high prevalence of missing data. Since most of the metabolic syndrome criteria (when adapted to include surrogate measures) are typically measured in primary care practice, this means that usual care electronic databases have the potential to be useful clinic population-wide to identify groups of patients at risk for diabetes and CHD. Analysis using individuals with all five criteria measured allowed us to compare our results to other reports and national data of prevalence of metabolic syndrome, but the algorithm using all patients with at least three criteria measured allows identification of a higher number of individuals with metabolic syndrome with high specificity.
Individuals with metabolic syndrome benefit from personalized lifestyle interventions to decrease metabolic abnormalities and prevent diabetes [18
]. One of the issues of primary prevention is how to identify patients with "prediabetes" since they are rarely aware of their condition and physicians seldom formally diagnose patients with metabolic syndrome [19
]. Our primary, more specific approach (≥ three criteria) allows case-finding for high risk patients for intensive lifestyle interventions to prevent diabetes. Alternatively, the more sensitive but less specific approach (≥ two criteria) identified a larger group of patients and could be useful for larger scale interventions such as targeted screening with information letters or invitations to group education sessions. Our cost data highlight the potential value to health care systems of metabolic syndrome detection for diabetes and CHD prevention.
Strengths and limitations
Strengths of this study include analysis of a very large number of patients, data from a primary care practice network representing real-world clinical care, and prospective follow-up of outcomes. We included all individuals using health care in the network, with no upper age limit, but only 1% of the patients were 85 years old or above at baseline, so this age range should not affect the main results. We identified diabetes and CHD in the EHR using a validated algorithm, so we are confident that the outcomes represent true incident cases. One limitation was the use of surrogate criteria when the formal criteria were not measured: our validation study showed that EHR metabolic syndrome (using both formal and surrogate criteria) had outstanding specificity (91%) for directly-assessed metabolic syndrome. Missing data were a concern and many of the patients in the At-risk-for metabolic syndrome group would probably fall into the category Metabolic syndrome if all five criteria would have been available for all patients. This limitation was addressed by an analysis of patients with all five criteria measured that confirmed our primary findings. Also, missing data and misclassification would likely reduce our ability to detect differences between groups, so our primary results probably underestimate actual effects.