The Role of Data Analytics in Reducing Low-Value Care
By: Neet Shah
Not everyone who undergoes a surgical procedure actually needs to have that surgery. In the medical space these occurrences – along with other procedures beyond surgical intervention – can be collected broadly under the label of “low-value care.” The title derives from the comparatively low value delivered to the patient for the associated costs, risks, and life complications incurred as a result of a given procedure.
Low-value care constitutes a non-trivial problem in healthcare; its continued occurrence is associated with increased possibility of patient complications, higher costs, and a rerouting of care delivery efforts away from higher value procedures.
Low Value to Patients
Joint replacement surgeries can sometimes be an example of low-value care. For the patient, these surgeries can sometimes greatly improve quality of life post-recovery, but often carry a considerable risk for post-surgical complications to negate any potential benefit. Moreover, by the common definitions of many insurance policies, they are more likely to be considered elective rather than critical surgeries. This typically causes them to fall outside of normal insurance coverage, making them prohibitively expensive for the average patient receiving a recommendation.
Sometimes candidates for these surgeries could benefit from more conservative treatments like physiotherapy and prescription medications; these methods are much safer for the patient with reduced risk of post-surgical complications like infection. Nonetheless, a good 20% of these patients will still opt for surgical intervention – an ultimately unnecessary avenue. 
Low Value to Providers
For providers, the costs of these procedures are eye opening. A study conducted in 2018 by Premier – polling 1100 hospitals over a nearly 2 year time span – indicated a total joint replacement surgery could cost as much as $30K depending on the hospital, although over half of these procedures cost between $12K and $18K.  But the risk of post-surgical complications does not only affect the patient, and perhaps represents an even deeper cost for providers – the cost of a harmed relationship between a patient and their caregiver.
Clearly, there is a strong case to be made for trying to find a way to reduce the number of these low-value care procedures where possible. The scale of this predicament certainly precludes any one solution, but even a method for reducing unnecessary surgeries could prove beneficial in the long run.
Pinpointing Pain Points
What if there was a way to make sure these surgeries are only recommended to people who truly need it? Individuals likely to receive a recommendation for this kind of procedure will usually fall into one of three categories: joint replacements such as knee or hip, or spine surgery, a category unto itself. A simple decision-making protocol could identify individuals at high risk of going in for surgery, which would then enable a nurse to talk to them before they approach a surgeon (Fig. 1).
Fig. 1 – Decision tree example for an arthroscopic surgery candidate
This simple act of intervention could be enough to prevent many “wrong” recommendations in the first place. Still, intervention is not prevention. The above decision tree is only useful if a patient presents with a doctor’s recommendation for chondroplasty of the knee.
How could anonymized patient data be used to develop an approach for identifying procedures likely to be considered low-value care?
What would such a modeling technique look like?
In order to be effective this approach would need a few key inputs:
- The right profile of the patient or population of patients receiving recommendations for surgery
- Overall population for the event in question (again using knee replacement as an example)
- Incidence rates for the event within the population (Fig. 2)
- Variable specifications for each event (Fig. 3)
Fig. 2 – Incidence rates for predicted surgical events
Fig. 3 – Sample variable specs for knee replacement event
Because the incidence rates are low, it is cost prohibitive to provide counselling to the entire population. This is where a logistic regression can help practitioners zero in on a smaller but higher risk population. The dependent variable for this model would include all individuals who would eventually go on to have a total knee replacement. A similar modeling approach can be implemented for other procedure types (Fig. 4).
Fig. 4 – Sample model performance summary
A broad application of this approach could yield considerable savings of time and financial resources for healthcare systems struggling with the ramifications of low-value care. Though not enough to solve the problem entirely, strategic application of a carefully considered data analytics model could enable a reduction in hospital hours spent on low value care procedures, better enabling providers to give the right level of care for more well-defined patient needs.
Interested in further applications of strategic data modeling in healthcare?
Will COVID Finally Give Big Banks Their...
What India's Infant Mortality Rate Tells Us...
The Seesawing Journey of Fintechs During...