Predictive modeling in healthcare: how to adopt it painlessly

Predictive modeling in healthcare: how to adopt it painlessly

January 8, 2020

Inga Shugalo

Healthcare Analyst

Healthcare providers spend most on the small segment of highly expensive patients. As stated in the article Focusing on High-Cost Patients — The Key to Addressing High Costs? by J. Michael McWilliams and Aaron L. Schwartz in The New England Journal of Medicine, nearly 75% of healthcare expenses are spent on 17% of the users. The main reason for this imbalance is the lack of continuous focus on patients with chronic conditions, which results in ill-timed interventions and increased readmission rates.

To reorganize care delivery in a more informed way, organizations turn to predictive analytics consulting to embrace the power of predictive modeling in healthcare. Additionally, there are complex algorithms at play, identifying processes and patterns invisible to the human eye, thereby helping researchers discover new medicine and treatment plans.

More and more healthcare organizations will be using predictive modeling in the future. According to Reports and Data, the global healthcare predictive analytics market was valued at $2.904 billion in 2018, and is estimated to reach $22.4 billion by 2026 at a CAGR of 29.8%.

However, while there is no shortage of needed data or custom healthcare software ready to tackle the challenge, the tough part is making this data actionable. For medical providers to do this, it is necessary to envision the full path ahead and be prepared to act in accordance with the received insights, such as dedicating a care management team to contact high-risk patients and their relatives to schedule a screening test.

What is predictive modeling in healthcare?

Predictive modeling uses data mining, machine learning, and statistics to identify patterns in data and recognize the chance of particular outcomes occurring. To build an accurate predictive model, developers first define the problem and collect data. For most analytical goals, a combination of clinical data and claims is used.

Modeling process

Each model is created with a range of predictor variables that affect future results. The model can use a simple linear equation or a complex neural network to perform advanced tasks. Data analysts run and assess different models to solve the defined problem. After the assessment, the model with the highest score is validated, tested, and run against a real-world dataset.

Here is an example of how predictive modeling assists physicians: a patient of certain race, age, medical history and genetics presents herself at a hospital with a newly diagnosed condition. The predictive model considers the new condition and all the patient-specific data mentioned above, while looking at the population cohort with similar characteristics to develop a treatment plan tailored to this particular patient. This plan is then presented to the attending physician for approval, exemplifying the use of big data in precision medicine.

Predictive modeling applications in healthcare

Employing predictive modeling helps doctors, support staff, and financial departments receive alerts about potential outcomes and risks, so that they are better prepared for the future. Below are some predictive modeling examples in healthcare.

Avoiding readmission

According to Medicare’s Hospital Readmission Reduction program, healthcare organizations are subject to penalties in case of patient readmission during 30 days after receiving care. Therefore, avoiding readmission will not only improve patient satisfaction but also save providers their money.

Predictive analytics can identify patients with a high risk of readmission after a particular treatment. Analytical tools can explain what exactly puts patients into this high-risk group, and guide on when to arrange for a follow-up and how to give optimal discharge instructions to avoid readmission.

Additionally, predictive analytics can incorporate social factors determining a patient’s health. When the social factors are considered, it becomes easier to understand when patients might not adhere to their prescribed care plans.

Having [social] challenges makes it much harder for people to be able to properly focus on their health. The science of predictive analytics is beginning to develop algorithms to track who these individuals may be and respond with carefully customized ways of reaching them to help them on their health journeys.

Rich Temple

Rich Temple

Vice President, Chief Information Officer at Deborah Heart and Lung Center

UnityPoint Health, a network of healthcare organizations based in Iowa, has reduced readmissions by 40% within 1.5 years of using predictive analytics. In one example, their home health team used predictive analytics to understand which patients became most vulnerable when the city was hit by a blizzard. This helped them focus their attention on particular individuals and reduce hospital visits.

Preventing suicide and self-harm

Identifying patients who are likely to harm themselves gives the opportunity to offer mental health care, which can hopefully prevent events of self-harm and suicide.

Kaiser Permanente together with the Mental Health Research Network used predictive modeling along with the EHR and a standard questionnaire on depression to identify patients who were at high risk of attempting suicide. They have discovered that suicide attempts were 200 times more likely among the 1% of patients flagged as high-risk by the algorithm. It should be noted that applying predictive modeling requires larger computational capacity, so the company should have performed EHR migration before implementing the technology.

Developing precision medicine and estimating a patient’s reaction

Medical researchers are turning to predictive modeling and analytics to supplement the traditional clinical processes. One interesting usage is adoption of predictive modeling for “in silico” testing. It involves importing an extensive number of patient-specific geometrics to a computer and using those for modeling the impact of therapies on patients. This is a promising way to evaluate new therapies without the need to recruit patients.

This method is now used in clinial trial management software for trial simulations related to degenerative conditions such as Alzheimer and Parkinson’s disease.

FDA’s Center for Drug Evaluation and Research (CDER) is currently using modeling and simulation to predict clinical outcomes, inform clinical trial designs, support evidence of effectiveness, optimize dosing, predict product safety, and evaluate potential adverse event mechanisms.

Scott Gottlieb

Scott Gottlieb

MD, Former Food and Drug Administration (FDA) Commissioner

Getting ready to embrace predictive modeling in healthcare

The HIMSS Analytics Adoption Model for Analytics Maturity (AMAM) and the Healthcare Analytics Adoption Model by HealthCatalyst are two similar models that introduce a step-by-step roadmap for adopting analytics at healthcare organizations.

Predictive analytics, together with its main techniques such as machine learning, data mining, and predictive modeling, stands at the next-to-last stage within both models. Accordingly, providers have to reach the needed rank of their analytical maturity with automated reporting, continuous data warehouse updates, and standardized patient records.

Healthcare analytics adoption model


Level 8

Personalized medicine & prescriptive analytics

Tailoring patient care based on population outcomes and genetic data. Fee-for-quality rewards for health maintenance

Level 7

Clinical risk intervention & predictive analytics

Organizational intervention processes are supported with predictive risk models. Fee-for-quality includes fixed per capita payment

Level 6

Population health management & suggestive analytics

Tailoring patient care based on population metrics. Fee-for-quality includes bundled per case payment

Level 5

Waste & care variability reduction

Reducing variability in care processes. Focusing on internal optimization and waste reduction

Level 4

Automated external reporting

Efficient, consistent production of reports and adaptability to changing requirements

Level 3

Automated internal reporting

Efficient, consistent production of reports and widespread availability in the organization

Level 2

Standartized vocabulary & patient registries

Relating and organizing the core data content

Level 1

Enterprise data warehouse

Collecting and integrating the core data content

Level 0

Fragmented point solutions

Inefficient, inconsistent versions of the truth. Cumbersome internal and external reporting

However, it is still possible to start incorporating predictive analytics and modeling from stage 4 in the AMAM and stage 5 in the Healthcare Analytics Adoption Model accordingly. These analytics adoption levels are mature enough and thus can help providers make accurate forecasts in the areas of waste and care variability reduction as well as in population health management.

Rule #1: Bigger isn’t always better

In healthcare, general-purpose features will lack utility and only introduce noise. So, while working with big data, organizations need to think small and narrow down their focus, targeting patient groups and particular care settings or departments within a hospital, not the entire health system.

For example, predicting the overall readmission rate could be useful for general reference. But in order to identify the right target for timely interventions and preventive actions, providers should tap into a specific condition and identify corresponding patient groups.

Rule #2: Know your data

Data comes from various sources, and those records are not always interoperable. So, it is important to understand the limitations and possibilities this heterogenous data brings into predictive modeling.

For example, episodic data with diagnoses and treatment plans tends to be unstructured, because some providers write examination notes on paper and then transfer them into EHRs as plain text. Others might just skip some fields. This data variability may significantly hinder further analysis.

At the same time, claims data is structured. It always includes ICD-10 and CPT codes, often identifying services and prescriptions that a patient received outside the health system or network. However, it isn’t precise enough to offer details on the patient’s health status changes within care cycles across different care settings. Also, claims come with a certain time lag by default, so they aren’t reliable for short-term predictions.

Predictive modeling in healthcare can fail where it is most valuable—in cases that occur outside medical facilities. These cases are usually related to chronic diseases and at-home post-surgery rehabilitation periods. The problem here is that post-discharge data can be unavailable or too fragmented because healthcare organizations either use patient-submitted surveys at specific time intervals or gather health status information during follow-up appointments and care episodes.

These fragments appear in EHRs and can point out only to severe health deteriorations, which may be useless for individual patients and patient groups due to many factors left undocumented in-between encounters.

To solve this problem, providers can turn to the internet of medical things and patient-generated data (PGHD) managed via mobile health solutions. This will help doctors not only make more informed and accurate predictions for individual patients, but also elicit patterns for different patient groups.

Rule #3: Implementation struggles are real

When deciding on whether to adopt predictive modeling, healthcare organizations should consider these three tiers of potential challenges: technology, security, and people.

Technology presents a pool of numerous issues when it comes to setting up clinical systems, workflows, and environments. Providers need to enable seamless integration between all their systems, including EHR, CRM, LIS, RIS and more, to ensure uninterrupted and correct data aggregation. The integration is also a critical step for enabling healthcare digital transformation across all provider’s facilities.

At the next stage, both structured and unstructured data needs to be leveraged, which implies further challenges associated with natural language processing and possible data errors. To make results immediately comprehensible for all stakeholders, data should be properly visualized, with summaries and automatically generated insights for each prediction.

Data privacy remains a concern, especially when it comes to collaboration between organizations or data exchange within one health system. Besides, population health management studies require data anonymization to be safely used by any organization.

To retain the utility of collected data for predictive modeling while complying with the HIPAA regulations, providers need to invest in information security assurance.

With expert assistance from vendors, providers can let go of many issues related to technology and security, but it will take a certain commitment from the organization itself to engage decision-makers and health specialists into predicting outcomes and acting on them.

Predictive modeling can’t be of practical use in vacuum. Health specialists are indispensable in this equation, because it is them who take part in the workflows, input data, and offer clinical experience to interpret predictions more accurately.

For the long-term success of predictive modeling at a healthcare organization, all the three tiers should be embraced by providers and technology enablers.

  • Technology
  • Security
  • People

3 myths surrounding predictive modeling

Predictive modeling is a relatively new technology, so consequently it is still ridden with misconceptions which healthcare organizations should understand and avoid.

Myth #1: Only large organizations can afford predictive modeling

Because of the big initial investment and the complexity of algorithms used, there is a tendency to believe that only smart hospitals of sufficient size have what it takes to initiate predictive modeling projects.

In reality, the determining factor is not the size of your organization. What is more important is how much data you have accumulated and can use to build predictive models.

Small healthcare originations that have been accumulating data for a year or more are also valid candidates for getting started on predictive modeling.

Myth #2: One predictive model is enough

Unfortunately, one predictive model will not be sufficient for satisfying all organizational needs and producing reliable results. Even a well-constructed predictive model can miss some valuable insights as well as produce false positives.

A more viable option is to develop several predictive models and run them in parallel. The point where all models intersect will form the option with the highest potential, and the point where most models intersect will be your second option, and so on.

Using several predictive models, you will save money in the long term, as you will be able to focus on more promising solutions and ignore false positives.

Myth #3: Building a predictive model is all you need to do

Adopting predictive models is not a one-time job. Those models need to be updated regularly to deliver optimal results. By comparing predictions to actual outcomes and understanding what went wrong, you will have grounds for adjusting the models.

Additionally, predictive models need to be updated when new data is added or new data sources are configured.

Despite challenges, predictive modeling is worth the effort

The value-based approach to healthcare encourages organizations to start using predictive analytics for developing personalized treatment plans, reducing readmissions, and generally improving health outcomes. There are additional benefits on the way, including more data-driven clinical workflows, optimized facility administration, more transparent supply chains, and cost-effective decisions.

However, predictive modeling adoption comes with its challenges. In addition to technology- and data-related challenges presented above, there is a moral challenge in identifying a reasonable point of human intervention. More precisely, predictive analytics predicts which treatment is most suited for a particular patient, as accurately as the algorithm can. But it is up to the human doctor to review those predictions and decide whether to act upon them or make an adjustment. Doctors should not just shift the responsibility to algorithms. They need to know when to interfere.

Despite all the challenges, though, the value delivered by predictive analytics is immense. This technology is worth your investment if you possess the data required for its operation, and if you are aware of the challenges and prepared to tackle them proactively.