As hospitals continue to work on reducing readmissions, another area of focus to reduce costs is through preventing potentially preventable hospitalizations, especially in chronic conditions. Potentially preventable hospitalizations, known as PPHs, are unplanned hospitalizations that have the potential to be avoided if timely and appropriate outpatient care had been received. However, in order to reduce these admissions, there has to be a means of identification. A number of methods have been reviewed to try and develop a way to identify those patients at risk of having a PPH.

In Australia, a Preventability Assessment Tool (PAT) was developed to attempt to identify patients at high risk of PPH The use of the tool compared to a similar assessment performed by an expert panel was assessed to learn if the tool identified appropriate patients. The  findings were recently published in a journal article. The expert panel consisted of a hospital physician, a primary care physician (or general practitioner (GP)), and a community nurse with expertise in the chronic conditions. The publication identified that the carefully constructed and developed PAT, when compared to the assessment of the expert panel, did not effectively identify those at risk of a PPH.

Another method to potentially identify these types of admissions is a hospital outreach program, also implemented in Australia. In the program, the patient record is flagged for areas of concern such as general health, medication, and wellness. Red flags are specific to disease or symptoms that have the potential for hospitalization. Trained telehealth guides reach out on a frequent basis (greater than weekly), while patients and caregivers can call in to the program at any time. Analysis of the flags being triggered through these phone calls may alert personnel to a deterioration in patient health, concerns about medications or a lack of support, and allow for outpatient care to be provided in a timely manner to avoid a hospitalization.

A study within the United States compared deep learning against a logistical regression model to identify prediction models for preventable hospitalizations, emergency department visits, and costs in heart failure patients. The study found that deep learning approaches identified these preventable areas more accurately than the traditional methods, indicating that outcomes are contributed to by clinical, demographic, and socioeconomic factors. The study found the main predictors for preventable hospitalizations in heart failure patients were diuretic usage, orthopedic surgery, and age (see Figure 1).