Is FICO-Scoring Patients Therapeutic?
The Fair Isaac Corporation recently announced the launch of FICO Medication Adherence Scores. FICO scores, which are famous for predicting whether an individual will become delinquent on a home or car loan, for the first time will be used to assess which patients are likely to take the drugs their doctors prescribe.
The FICO Adherence Score is an algorithm that FICO developed based on the close study of almost 600,000 patients suffering from asthma, diabetes, and heart disease. FICO identified which patients were likely to have their prescriptions filled and re-filled. FICO then used data mining techniques to identify correlations between prescription filling and consumer information already in its credit history databases. This information, combined with data gleaned from a patient’s own history of getting prescriptions filled, could predict patient behavior. As it turns out, individuals who rent their homes, live alone, don’t own cars, or have started a new job recently are less likely to follow their doctors’ advice. The risk factors that predict a loan default and a failure to take Advair are not identical, but there is evidently some overlap.
Research cited by Fair Isaac suggests that noncompliance with drug treatment regimens cost the American health care system some $250 to $300 billion per year, approximately thirteen percent of total health care spending. Despite the significant health benefits from a system that might help doctors and insurers identify noncompliant patients who would benefit from reminders to take their medicine and follow-up nurse visits, the Medical Adherence Scores sound frighteningly Orwellian and Kafkaesque. Critics raise concerns about patient privacy and the unreliability of FICO scoring in general. They rightly note that patients are people, not automatons, which means even the best algorithms will make mistakes. Patients without cars or roommates have wondered whether they might face discriminatory treatment and whether the Medical Adherence Scores would be used to set insurance premiums. (Patients who do not get their prescriptions filled regularly may actually see their health insurance premiums decline, at least in the short run, but they could see their life insurance premiums rise.)
Alas, it isn't appropriate for FICO's critics to dismiss Medical Adherence Scores by comparing our new reality to a perfectly virtuous world. A ban on the use of FICO scoring in medicine wouldn’t eliminate a common dilemma: The best treatment plan for, say, congestive heart failure, may require vigilant follow-through by the patient. But if such compliance is unlikely, the optimal treatment may be another therapy altogether. Organ transplants represent a particularly stark choice. Transplants have great potential to improve the lives of recipients, but a lack of follow-through by a patient and her caregivers may expose the recipient to life-threatening risks and result in the waste of a very precious resource that could have saved another person’s life.
A physician must have some criteria for deciding which type of patient she is treating. The patients themselves are not always reliable sources for this screening. Few patients will admit to their doctors (or to themselves) that they are unlikely to follow through.
When physicians do not know a patient well, they sometimes rely on proxies that are more distasteful than car ownership in assessing the odds of follow-through. Some physicians rely on the equivalent of old wives’ tales. But as I detail in chapter eight of my brand new book, Information and Exclusion, recent research on health disparities suggests that junk science decisionmaking may be the least of our worries. One study in the American Journal of Transplantation identified a greater propensity among nephrologists to refer children from affluent families to transplant surgeons. The physicians assumed that wealthier parents would be more likely to comply with rigorous postoperative recovery protocols. A separate study in Social Science & Medicine found that the physicians surveyed viewed African Americans as less likely to comply with treatment regimens. Such racial profiling by physicians may contribute to disturbing phenomena like doctors’ tendency to prescribe narcotic pain medication far more readily to Caucasians than African Americans.
Physicians are not going to treat all patients equally, and maybe that is for the best. We want compliant and noncompliant patients alike to get the respective treatments that will be most therapeutic. But deciding who fits into which group can be a daunting challenge, particularly for specialists with large practices and a population of patients who have bounced from doctor to doctor. To be sure, FICO’s Medical Adherence Scores are imperfect. We know that many errors in consumer credit databases go undiscovered, and getting even acknowledged errors fixed can require substantial perseverance. But at least FICO’s predictions will be based on hard data that patients can access, and the law can ensure that factors like race and national origin are not used as inputs into the algorithm.
Fair Isaac Corporation is the first entrant into this market, but they should face competitive pressures to improve the accuracy of their scoring as time passes and patient behavior patterns change. The biases that some physicians rely on are stubborn, unscientific, difficult to detect, and far more disturbing alternatives for predicting patients’ behavior.