FAQ
Biometric Health Frequently Asked Questions
Q. How is personal data protected?
A. BMH will comply with all mandated requirements, for example, data will be encrypted in transit and in rest.
Q. Where is the de-identification process best implemented?
A. De-identification is best implemented on the client's infrastructure. Identified information however, is necessary for applications such as care management.
Q. How is data access determined?
A. Those with business and/or clinical justifications for data access the data will be granted an ‘only the minimum required’ privilege. These privileges will vary based on level required, algorithm development v. statistical modeling/validation v. data mining v. reporting, etc.
Q. How does the data integrate with my systems?
A. Data transmitted via a standard structure file (JSON, Restful API) or 'delimited' files (comma, tab, etc.) either scheduled or real-time.
Q. Who owns the data?
A. Historically, the customer owns the data.
Q. Who has access to the data?
A. The customer will grant access to those persons/entities providing appropriate justifications required of said data.
Q. Is patient consent required?
A. At present, patient consent is not required; however, certain BMH products/services in development will require some level of patient consent.
Q. How do we transfer the data initially? Do we need access to their systems?
A. The most efficient path is for the customer to provide us the data in a number of potential formats (see above); however, customers can also grant BMH direct access to their systems/data warehouse, etc. should they desire that approach.
Q. How is an ROI calculated?
A. ROI calculation is scenario-driven. Generally, there exists both quantifiable (something happens by design of intervention, e.g., routine mammograms, diabetic podiatry and retinal exams, closed gaps in care); and, unquantifiable (something of high probability is avoided).
The ROI is calculated as premiums paid v. quantifiable and unquantifiable returns. The latter calculation is complex and requires factoring the determined 'savings' by the probability of predicted event happening.
Q. What is special about BMH solutions? Why is it different from other solutions?
A. BMH develops standard predictive modeling packages; however, within those standard algorithms/models exists all customization required to produce powerful, accurate predictions.
BMH also offers moderate (rare event) to highly complex (behavioral prediction) models as well as custom, 'boutique' products/services that are developed to answer any business question outside of the standard, moderate and complex package portfolios.
Q. What is the benefit of the BMH philosophy/approach?
A. Our BMH models cannot be purchased off the shelf; therefore, providing the customer a distinct competitive advantage.
Q. How should the data be structured?
A. Consistently structure data certainly brings an efficiency to these advanced analytic solutions. That said, it is critical to note that BMH can insource and format nearly any structured data to meet our internal requirements for data quality and sufficiency as well as optimize algorithm/model performance.
For example, we could model against data provided in an excel spreadsheet or workbook with numerous worksheets, a delimited file, clinician notes, etc. BMH has also developed some language processing routines within SAS that can identify and exact specific symbols, biometric results, keywords and phrases.
Future offerings will also contain ‘context’ recognition.
Q. Is my data quality ‘poor’?
A. This is a very subtle yet complex question. There are many definitions of ‘poor’ data quality; including but not limited to:
- Volume too low or lacking statistical power (variance) to find statistically significant results, if said results are to be found. Each metric/measure is first tested for statistical power; absent this ‘power’ results will be presented as 'descriptive.
- Volume too low to model at all. This does not happen often. BMH has developed methods to deal with very low volume datasets. The least case scenario would be to characterize these data and apply mining techniques to extract that which can confidently be acted upon.
- Sufficient volume but values within fields that are invalid (e.g., blood pressure values outside of established clinical ranges), missing values, etc. BMH will implement 'proxies' whenever possible, to confidently fill in information that is missing using valid information that is present on the same records.
Q. How does BMH determine data quality? Sufficiency? Does BMH provide a roadmap for data quality improvement?
A. BMH performs rigorous data quality and sufficiency testing prior to every algorithm and/or model development. This 'stress' testing of the data results in clear emerging patterns of improvement opportunities that will be communicated to the customer along with our specific recommendations.
Q. How does BMH validate models and which statistical methods are most commonly used?
#1. How are the models validated? A. The models are developed using a split halves or split thirds random population approach with a 'hold-out sample' (persons not involved in the model build) to test the power and accuracy of the model against an ‘outside’ population. BMH also has strict c-statistic and Adjusted R2 requirements (as well as many other statistical outputs) to help determine the best, or, 'champion' model. BMH will also determine selectivity and sensitivity by publishing a 'confusion' matrix (i.e., true v. false positives, true v. false negatives).
#2. Which statistical methods are most commonly used? A. BMH utilizes only industry proven statistical methods; each time starting with the most simplistic approach (linear and non-linear models) then advances in complexity until such time to optimal model/models is/are found. This approach, when appropriate, has a two-fold benefit.
- Customers who conceptually understand any given methodology presented tend to be more willing to take confident action against those model results.
- Simpler equation formulae can be built into our BMH microsimulation portals, allowing clinicians, underwriters, finance leaders, FWA investigators and the patient themselves to ‘interact’ with the model, providing real time results based on what characteristics are input.
Q. Does BMH offer machine learning/AI capabilities? Resources?
A. While SAS has been a component of machine learning and AI solutions for the past decade, BMH currently utilizes platforms such as Azure, AWS and Google that contain machine learning and AI. BMH also has experience with Neural Networks.