- Self-reported patient data on health conditions, status, and utilization may be the key to identifying high-cost patients and guiding them to care management models to reduce their spending, a recent American Journal of Managed Care study indicated.
As providers assume financial risk under value-based purchasing models, they should implement processes to identify high-cost patients and prevent unnecessary or excessive spending on the patient population.
However, providers may face challenges with obtaining the data necessary to pinpoint patients who may incur significant healthcare costs. Providers tend to rely on claims and EHR systems to gather the required information, but they may not have access to this data source for patients enrolled in private plans.
Researchers noted that about 13 percent of privately insured individuals switch their health plan within a year and approximately 9 percent change their usual physicians, a cited National Institute for Health Care Reform study found.
The Affordable Care Act also propelled more individuals to switch health plans and providers as additional health coverage options became available under health insurance exchanges.
Providers also experience difficulties accessing claims and medical records on new patients enrolled in private plans because the information is generally not available to providers when they accept new patients.
But health needs assessments and health risk appraisals may be the answer to limited data access, the study showed. The analysis of health surveys from 8,624 hourly wage workers at Chrysler, Ford, and General Motors with private insurance revealed that self-reported patient data in the absence of EHR information was a reliable indicator of if a patient would be considered high-cost.
Researchers examined six high-cost patient identification models. Model 1 included claims data in conjunction with self-reported information on age, gender, and race/ethnicity, while Model 6 included all patient-reported information and claims data.
Model 2 started with just self-reported data on demographic and socioeconomic measures and each subsequent model added survey answers on chronic conditions (Model 3), health status and behavior (Model 4), and hospital inpatient and emergency department use in the past year (Model 5).
While Models 1 and 6 were the best predictors of high-cost patients, researchers found that incorporating self-reported patient data increased a model’s predictive ability compared to high-cost patient identification models that just used demographics and socioeconomic status.
For example, Models 1 and 6 had C statistic values of 0.78 and 0.8, respectively. Models with a C statistic of 0.7 or higher are considered to be good predictors of high-cost patients.
The models that only used patient-reported data started at a C statistic value of 0.57 under Model 2, which only included demographic and socioeconomic information. As the models incorporated additional self-reported data, the C statistic value rose to 0.70 for Model 3, 0.72 for Model 4, and 0.73 for Model 3.
However, a sensitivity analysis that measured the percentage of individuals who had high healthcare costs and who were accurately predicted by the models uncovered that few high-cost cases were correctly pinpointed by the best three models (Models 1, 5, and 6) based on the 75th percentile threshold.
Model 1 had a sensitivity value of 45.6 based on the 50th percentile threshold. However, the value fell to just 3.4 under the 75th percentile threshold.
Similarly, Models 5 and 6 had sensitivity scores of 20.9 and 41.6, respectively, at the 50th percentile threshold, but the values dropped to 2.9 and 11.1, respectively, once the threshold increased.
The low sensitivity scores indicated that the models did not identify a large share of patients who could potentially benefit from receiving care management services. Therefore, providers would not be able to reduce the healthcare costs associated with this population.
On the other hand, the three models scored high specificity values based on both percentile thresholds. The specificity analysis showed that the share of non-high-cost cases that were accurately identified.
With scores over 90, the models successfully predicted non-high-cost patients, suggesting that they would prevent providers from engaging these patients in expensive care management or other specialized services and wasting resources.
Despite model shortcomings under the sensitivity analysis, researchers concluded that patient-reported data can be a useful source of information for providers looking to find their high-cost patient population.
They also noted that patient-reported data is less susceptible to upcoding of claims and EHR data. Providers paid according to patient risk scores may code patients as if they are sicker in order to maximize their revenue. But upcoding could skew the identification of truly high-cost patients.