(Note: Caliper is a fully owned brand of Foresight Health Solutions LLC)
You have heard the saying, if you have seen one Health Center (HC), then you have seen one HC. The vast differences between HCs have many reasons, starting to a large degree from different populations, different leadership and approaches to care, different local legal and community support. The interwoven web of local challenges to the HC population, the variety of community services such as food programs, events such as a fire in a community shelter or an influx of homeless due to temporary drug legalization all contribute to a very custom contextual situational environment. Nationwide models that generalize and average over such diverse situations cannot capture its intricacies.
Key to a technology’s relevant operationalization is a careful consideration of the specific situational environment. Foresight Health Solutions brings dedicated clinical expertise to the table, some of us have been working in healthcare for three decades, others still regularly work actively in patient care. Only this up-to-date knowledge of the care management-specific problem and clinical workflows allows for unparalleled relevance of the actionable results our analytics produce.
Caliper as Your Partner
When it comes to selecting your partner for your journey with data-driven decision making, for exploring or expanding your use of AI and machine learning to aid critical planning and care decisions, you need someone who understands your needs and desires, your patient population and your staff, your processes and constraints. Foresight Health Solutions is a company that specializes in AI and data science for Community Health Centers - this is all we do.
For us, AI is not a tag-on, it is not a recent addition to our marketing language or a quick slap-on tool. AI, to include machine learning and data science and analytics, are at the core of our business. We were founded in 2019 on the premise that AI can improve healthcare especially for the underserved. AI has become a household term since November 2022 when ChatGPT was released to the public, and many companies are now marketing an “AI functionality” but rarely does this mean there is a dedicated team that understands the technology as well as the domain - CHCs.
Our organization is staffed with experienced data scientists who work very closely with experienced CHC and care experts. We have medical doctors, tech MS and PhDs, all working together on the same models and products. This makes our products uniquely relevant and easier to operationalize than the typical products from larger, more generic and less focused technology companies that are only recently attempting to sell to CHCs.
Comprehensive Data for Accuracy
Consider as many factors as possible. Recognizing the importance of a holistic understanding of a patient’s situation, Caliper integrates a comprehensive set of patient data encompassing diverse demographics, medical histories, treatment regimens, and Social Determinants of Health (SDoH). The Caliper comprehensive risk model captures many more variables than traditional models and other machine-learning based solutions, with a focus on specific HC programs and intricacies that can only be discovered in close collaboration with HC staff. This amount of data sources and variables can only be tackled with machine learning as traditional data science methods will not capture such sparse correlations. This comprehensive data approach ensures a holistic assessment of each patient's health status and unique needs.
By capturing intricate relationships and interactions within the data, this model achieves a higher level of accuracy in forecasting healthcare costs. Compared to traditional and other AI-based models, the FHS models:
- Integrate Over 130 Dimensions: The FHS models can incorporate more than 130 dimensions of medical, social, demographic, and environmental risks, compared to the 80-100 variables used by the Johns Hopkins Adjusted Clinical Groups (ACG) System and other commercial models, which are limited to historical data, claims, and demographics without including Social Determinants of Health (SDoH).
- Customization for Community Health Centers (CHCs): These models are tailored to the populations served by community health centers, which face higher risks than the average Medicaid population. FHS models compare individual patients to the average risk of CHC populations, rather than the general population, which is typically used by other risk models for relative risk identification.
- Access to Diverse Data Sources: This customization is facilitated by our close relationship with CHCs and their payers, providing access to more data sources than a single-source solution can utilize, such as AI solutions that only access claims data or electronic health records (EHR) data.
- Comprehensive Risk Analytics: The FHS models combine comprehensive risk analytics with a distinct SDoH score/disparity index and an impactability score, using actual care management services data to measure impact. Recently, we have added a quality risk index/score to the model.
- Simulation and Prescriptive Analytics: The models can simulate interventions at both the patient and population levels to predict patient outcomes and cost impact. They also use prescriptive analytics to recommend care-enabling service interventions.
A second quantity needs to complement an accurate cost-based risk prediction: a prediction of the degree of opportunity of positively impacting the patient’s healthcare outcomes. Caliper performs a secondary analysis to determine the impactability of the patients. This is very important for operationalizing the results, as purely high-risk/high-cost patients might not be impactable at all. For example, patients to which the HC has repeatedly reached out without any response from the patient would be considered with a low impactability score as they are unlikely to benefit from additional care-enabling services or interventions.
This combination of features and functionalities makes our system much more complete and presents a unique set of tools for accomplishing community healthcare goals.
The actual models are the result of a very comprehensive search over many staples of ML algorithms and contemporary algorithms including Deep Learning. Deep Neural Networks, in particular, give the models sufficient flexibility to uncover intricate patterns within vast datasets, while avoiding overfitting to training set peculiarities. This capability enables precise predictions for healthcare risk, ultimately leading to improved care outcomes and optimized resource allocation.
Robustness and Missing Data
AI-based risk scores also outperform classical regression models and especially simple linear models in cases of missing data, which is a common occurrence with highly multivariate problems - the kind of problem settings posed with tens or hundreds of possible contributors to health.
Quantitative Evaluation
The CMS HCC Model has exhibited significant limitations in cost prediction, substantially underestimating costs with an average risk score of 0.49. This means the model only accounts for half of the total cost for a given population. In contrast, Caliper’s Comprehensive Risk Model predicts more precise and more accurate cost estimates for patients.
Quantitatively, the accuracy of predicted numerical values is measured with the R-Squared metric which compares the predicted with the actual quantity.
R2 (“R-squared”) value is consistently higher than the CMS HCC model. In essence, our model has an R2 value that's 3x as high as the unmodified CMS model (3x as accurate), and 42% better if the CMS model is modified (additive shifted) to predict costs with a mean that matches last year's mean costs. This remarkable improvement of the Caliper Comprehensive Risk Model highlights the transformative power of our AI-driven approach to risk assessment.
Integral Natural Language Processing (NLP)
The quickly-increasing capabilities of Natural Language Processing (NLP) have been integrated into many of our solutions, and some standalone modules. This deep integration harnesses the insights that this technology can generate for extraction of structured information from free-text notes and remarks, it improves the comprehensive knowledge of patients contributors to health, it captures additional interventions that otherwise go unrecorded. Depending on the quality of structured coding of SDoH and care-enabling services, FHS’ language processing solution can uncover between 5% to over 600% more SDoH and interventions than were coded in structured data. Historically, the average is around 40% of additional information that would remain locked away in unstructured, natural language in EMR notes. This additional information presents a more complete patient picture and in some cases even permits higher reimbursement rates.
The Caliper data extraction from written notes is more customized than what the more generic, larger competitors can offer. Our process relies on NLP with several steps, some of which are customized for the terminology and customs of the organization and their patient characteristics, their care programs, local community resources, and other regional factors.
Augment the Human Workflows, Support Human Decisions
Consider existing workflows and specific decisions that can be supported with data insights. The goal is not for technology or AI to replace the human, the goal is to augment the human with data-driven insights, such as a longer time history analysis (longitudinal data) of the patient data, consideration of more factors (breadth), integration of multi-factor correlations (connective fabric).
A key factor is relevance and usability of the results. The performance measures of the FHS Comprehensive Risk Score have translated into tangible improvements in health outcomes, and have, for example, resulted in 13% in care expenditure savings over just one year. FHS works with our customers to identify a usable and convenient way to operationalize the results from scoring analysis, whether through deep integration with a care management system or through universally readable Excel sheets.
Rapid Iteration and Integration of Feedback
Our Comprehensive Risk model is customized for every organization and their population’s characteristics to predict the most accurate outcomes possible. For example, some of the models are cost-based, and costs can vary dramatically according to HC location. Without accounting for these differences, models are going to return less appropriate results as we have seen many times with AI modules from some of the larger companies.
Our regular cadence of connecting with customers lets our dedicated staff take feedback into account quickly and iterate on outcomes, UI functionality, prescriptive models, and all our services. Facilitating this is what we call a Rapid Iteration and Model Generation (RIMG) functionality of our core infrastructure. This lays the foundation for white-glove service to your CHC or care organization.
These rapid response times allow us to be a step ahead for the competition, to anticipate needs and to prepare tailored solutions.
Operationalization and HIT Integrations
Tailored solutions are key to quick operationalization. Some of our products and visual outputs conceptually sit on top of Business Intelligence (BI) tools such as Tableau or PowerBI but they are ready to use out of the box, they are customized with more relevant information, they have appropriate terminology throughout, and they have been curated for and reviewed with users.
Because these solutions show domain-specific information, because they have been prepared and are ready to use, you don’t need to hire a data analyst to wrangle your data. FHS solutions are several steps closer to operationalization: the more directly usable the result, the easier it is to operationalize. Another key aspect is our solution integration into other software.
Integration Into CMS, EHR, and Other HIT Systems
The FHS analytics and modules are readily available as a standalone web-application or they can be accessed through a variety of Health Information-Technology (HIT) systems such as Care Management Software (CMS), Electronic Health/Medical Record systems (EHR/EMR), solutions for medical scribes, or your CHC’s custom software. We have integrations into the major EHRs, are a Preferred Partner of OCHIN Epic, and have automated data analysis or integrations with Azara, CCS, FirstPath, and several other systems through Single Sign-On (SSO), SMART on FHIR, or other standardized authorization protocols.
Response time: most results are pre-computed and stored in a sub-second access time database. To allow for maximum flexibility for users of the FHS UI, some custom results are calculated on the fly and might take a few seconds, such as statistics on complex filtering operations on 100s of thousands of patients.
Interested to learn more? Reach out to us info@caliper.care.
