With more statistically rigorous analytic methods to further automate insight identification, data science and machine learning can help health systems align effective measures with specific improvement goals more accurately and faster than typical data analysis.
This article explains how improving healthcare with data science can save organizations time and money by targeting actions that will help them reach their goals, while avoiding spending resources on measures less likely to lead to desired outcomes.
As part of an initiative to reduce readmissions in its orthopedic surgery population, a large health system proposed preoperative optimization measures for each patient based on that individual’s circumstances. For example, one measure might recommend delaying total joint replacement surgery for a severely obese patient until their BMI was within a range often associated with better postoperative outcomes. To help patients achieve this goal BMI range before orthopedic surgery, severely obese patients could undergo bariatric surgery, a procedure to make the stomach smaller.
However, when this health system used data science analysis (specifically, a logistic regression model) to test its hypothesis about BMI and rate of readmissions, it found that BMI was not associated with lower readmissions. The regression model insight showed the health system improvement team that an intervention (e.g., bariatric surgery) to reduce BMI before orthopedic surgery may not be an effective readmission improvement measure and that the system could better allocate its resources to meet its goals. For this specific analysis, data scientists elected to use a statistical model instead of a machine learning model due to the different purpose each serves (i.e., statistical models are designed to understand relationships between variables, whereas machine learning models are designed to make the most accurate predictions possible).
There are four key reasons data science helps health systems better align measures with their improvement goals and maximize their ROI:
To ensure that the health system’s preoperative optimization measures were statistically associated with its target outcomes (e.g., that patients undergoing total joint replacement with BMIs within a specific range had lower incidence of readmission), the improvement team worked with data scientists to develop a logistic regression model.
The goal of the regression model was to understand how the risk factors from the proposed optimization measures were associated with readmissions, as well as to control for other patient-specific attributes—rigorous improvement over more simple methods—to better understand how each factor affected the target outcome.
Data scientists used several control variables for the regression model:
After accounting for the risk factors described above, the output from the model showed that patients with a behavior disorder, low hemoglobin, renal disease, and those taking opioids before orthopedic surgery or who were male tended to fare worse in regards to readmissions within the health system’s orthopedic population. Conversely, patients undergoing knee procedures (compared with hip procedures) and those who reported using alcohol tended to fare better.
The analysis, however, showed that some of the factors the improvement team expected to be associated with readmissions did not have a statistically significant association. For example, a higher BMI was not statistically significantly associated with readmissions, and the magnitude of the association was approximately zero. This implied that bariatric surgery wasn’t an effective way to prevent readmissions among patients undergoing orthopedic surgery, as the surgery did not impact the likelihood that an obese patient would return to the hospital. If these patients were to undergo bariatric surgery solely to reduce their risk of readmission, they might be going through an unnecessary procedure, including its associated costs and risks.
To successfully impact orthopedic surgery readmissions, the data scientists and improvement team needed to identify factors strongly statistically associated with readmissions and determine whether they could affect change around those factors.
The improvement team worked through the factors with strong associations with improvement to assess whether they could improve an existing process or determine if there was a new process they had the resources to implement:
Additional regression models provided other balance measures (outcome measures that may be important but aren’t the focus of the readmissions project) the same level of attention in the statistical analysis as readmission measures. The analysis showed how optimizing a process for readmissions might also help or hinder outcomes including length of stay (LOS), mortality, or cost.
Reducing preoperative BMI may be an effective measure for goals other than reducing readmissions. For example, statistical modeling showed that a high BMI was associated with a longer LOS following orthopedic surgery. Though bariatric surgery appeared to be an ineffective measure for reducing readmissions, it may be an effective action if reducing LOS is an important improvement goal.
To sustain improvement work, organizations must consider if the effort invested in improving specific processes will yield a worthwhile ROI. Otherwise, their energy and resources are better spent on other outcomes.
When data didn’t link patient BMI to readmission rates, the improvement team assessed other preoperative factors that might identify patients more likely to be readmitted. Looking retrospectively at reasons for readmitting helped the team understand why patients were returning.
Backed up by data and rigorous analysis, the team determined that, given that most orthopedic readmissions were unrelated to surgery, a systemwide approach to readmission reduction would be more effective than implementing process improvements by department (e.g., orthopedics). The team further supported this conclusion after assessing the potential impact and volume behind specific preoperative interventions, ultimately changing its improvement focus to systemwide data science-driven initiatives (e.g., opportunities to reduce cost, variation, and waste, and to improve efficiency).
When the organization followed the data and used more advanced data science methods to evaluate its optimization criteria, it found that delaying orthopedic surgery based on a patient’s specific preoperative attributes (e.g., BMI) may not be an effective measure for avoiding readmissions. The improvement team learned that many of the factors associated with increased readmissions do not tie to processes they can control directly. The team also learned that readmission processes it can impact may only require only a light touch (e.g., behavior disorders and opioid use) but yield meaningful improvement and favorable ROI.
Health systems can decrease the risk, cost, and time associated with outcomes improvement and accelerate the process by using data science to help determine which measures will help them meet their goals for their specific populations. As the industry continues to work toward outcomes improvement, leading organizations will rely on data science and machine learning to test hypotheses, identify opportunities faster and more accurately, and ensure that their improvement measures support their overall goals.
Would you like to learn more about this topic? Here are some articles we suggest:
Would you like to use or share these concepts? Download this presentation highlighting the key main points.