By using predictive analytics as well as risk stratification techniques, healthcare organizations have found they can reduce the number of chemotherapy-related hospitalizations.
The healthcare industry continues to find ways to use data analytics to improve patient outcomes and make organizational operations more efficient.
In recent years, one area has been in treatment of cancer patients.
The predictive analytics strategy has been used by some hospitals to reduce overall patient readmissions. By using predicative analytical modeling, patients with a likelihood of returning to the hospital – based on factors such as age and type of condition – are flagged and intervention steps are taken.
Those steps often include more frequent communication with patients. In some cases it even means the appropriate medical professionals make home visits.
The strategy for cancer-related issues follows this same course. It not only improves patient outcomes, but also opens a new area for those with data analytics degrees to put their skills to use.
Predictive modeling essentially assesses the probability of a future event based on input variables. In healthcare, those variables can include a wide variety of data points. Some examples of often used input variables include age, medical history, frequency of doctor and hospital visits, and clinical decisions made by medical professionals. The variables are combined with the overall data on those with conditions such as a cancer.
Medical professionals outlined much of the strategy during the 2017 Managed Care and Specialty Pharmacy annual meeting in Denver. Leaders from medical provider organizations and insurers spoke about the current use and future potential of predictive analytics and risk stratification.
Predictive modeling is also used by researchers, such as those at the University of Pennsylvania Perelman School of Medicine. They were able to determine which lung cancer patients were at a higher risk of needing treatment in an emergency department.
In the area of cancer, every case is considered high risk. However, using clinical data, analysts can identify and assess more finely grained details on the likelihood of certain events given each patient’s condition and medical history.
It’s essential that healthcare organizations define goals before beginning research using data sets. Reducing unnecessary costs for patients can be one such goal. Reducing the number of visits to the hospital can be another. Overall, the goal behind all data-related projects is to improve patient health outcomes.
Three Stages of Predictive Analytics
Typically, the use of predictive analytics breaks down into three stages.
First, a predictive model is built based on the data. This usually requires a large amount of data. According to remarks at the Denver convention made by Mir T. Mimazari, assistance vice president at New Century Health, data from at least 1,000 patients is needed.
The model is based on trends found within the data on patient outcomes and costs. The model is then tested. Data is kept to see how well the model predicted patient outcomes and costs. If the model is found sufficient, it can then be moved into the validation stage.
In validation, the model is tested against new data over a set period of time. If it’s accurate, then the findings can be combined with clinical treatment and decisions by medical professionals. The success of those initiatives is also tracked and changes are made as the data indicates is necessary.
In another words, as Mimazari said at the conference, it creates “a continuous learning loop that leads to better cancer care delivery.”
He later added, “It’s not magic.”
Part of the predictive analytics and modeling process involves risk stratification. It differs from the data used in modeling. Most of that data is taken from claims filed with insurance companies and patient records. However, some of this data can be months old.
In risk stratification, clinical data on what happens during treatment is collected. While not as detailed as claims data, it allows for faster action once correlations between clinical decisions and patient outcomes are found.
For example, this data can provide a quick way to identify patients who are likely to end up in the emergency room. Such visits are very predictive of future inpatient care and hospital visits. Other factors include patient age, the stage of cancer, the type of chemotherapy treatment being used and changes in a patient’s weight.
All of these factors help to calculate a “risk score” for the patient. Medical personnel can consult this score to help guide clinical decisions.
These examples show the potential of data analytics in healthcare and how it can improve patient outcomes and reduce costs. As with most areas of the field, the strategies used are expected to change over time. They will become more detailed and streamlined as new data shows the impact of predictive analytics and modeling.