There is a large volume of multi-variable cross-functional healthcare data out there, which if properly integrated, could be used in the development and on-going running of some very interesting and powerful advanced analytical models. However, the application of advanced analytics in the healthcare sector is still in relative infancy. In this post I explore how predictive analytics, segmentation and text analytics can be utilised in healthcare.
Every day, providers (hospitals, clinics, doctors, dentists, chiropractors, radiologists, etc), ‘payer’ organisations (i.e. healthcare insurance organisations), life sciences organisations, as well as the developers and marketers of pharmaceutical products and medical devices capture and generate a vast amount of detailed data. Sadly, only a very small portion of this goldmine of data is ever analysed in depth.
Predictive analytics
The use of predictive analytics is best illustrated by a case study on medication adherence [1]. Tens of millions of patients in the United States are living with serious, chronic diseases. Nearly half of all medication prescriptions are not followed properly, resulting in an estimated 125,000 premature deaths in the United States each year [2]. A 2011 study showed that a non-adherent patient with high blood pressure spends an average of $3,908 more per year for healthcare than an adherent patient. For congestive heart failure, the additional patient cost is estimated to be $7,823, and for diabetes it is $3,765 [3].
Current approaches to combat non-adherence often do too little, too late. In many cases, the doctors only become aware of the negative consequences after patients have stopped taking their medications. Above all else, healthcare providers have a moral and ethical imperative to do what is best and safest for their patients, and in many instances that includes trying to determine which of their patients are heading for non-adherence.
The FICO Medication Adherence Score was developed to alert doctors, pharmacists and health plans of patients most likely not to adhere to their schedules, and who are the most vulnerable to negative outcomes. These scores are determined by analysing publicly available data such as the patient’s age, gender, marital status, geographic region and type of disease. Properly analysed, the resulting scores indicate the probability of patients adhering to their prescriptions during the first year of therapy.
Healthcare providers cannot possibly pay attention to every patient. However, with a risk score that accurately highlights the probability of medication non-adherence; the most vulnerable patients can be targeted with intervention tactics most applicable to their particular needs.
This is but one use of predictive analytics in healthcare. Health insurers can determine their members’ lifetime value, a measure often accurately predicted in telecommunications and insurance businesses, to calculate fund and option profitability. Other measures that can be predicted include claim patterns, patient arrival rates, hospital bed occupancy, likelihood of policy/plan lapses, occurrences of complications or secondary admittances, and so on.
Another massively beneficial utilisation of predictive analysis is in combating healthcare fraud. Estimates of annual losses to fraud are high, but the actual losses are way higher, ranging from 3% to 5% of national healthcare expenditures. The truly scary statistic? Credit card fraud accounts for 100 times less than healthcare fraud [5].
Predictive models must be deployed that use historical claims and other claims-related data to predict the risk of the current behaviour or claim being assessed. Profiling technology is also deployed to compress terabytes of claim-related data into essential informational packets. The technology is dynamically learning and the system is therefore able to detect fraud with increasingly greater accuracy over time. This gives organisations a greater opportunity to catch out fraud, abuse and erroneous claims prior to payment, before losses are even incurred.
Segmentation
Not all people react and behave the same, therefore it is necessary to segment the member, patient and provider bases along with the individual’s data. There are usually some well-identifiable groups that behave similarly and by accurately identifying the criteria that distinguish the members of the respective groups, and then adjusting interactions to suit the nature of the group, much more appropriate interactions, with much more positive outcomes can be obtained.
As in all other industries, the primary focus is currently still on the customer – in this case the patient or the member of the medical scheme. As long as the segmentation models are accurate and effective, patients and members can be analysed successfully and therefore be targeted suitably in the context of their segments. With segmentation, the effectiveness of the model must always be analysed and the model must be adjusted when necessary.
However, the area that is still greatly neglected in the healthcare industry is the segmentation of providers. A large healthcare insurer typically has high numbers of thousands of providers on their books, with which they have individual or consortium-based contracts. Segmenting the providers based on performance, patient satisfaction, adherence to scheme rules, compliance, profitability, etc., can lead to a much more appropriate treatment of providers in their respective segments.
Text analytics
According to Seth Grimes, President of Alta Plana Corporation and author for InformationWeek.com, text analytics is another application of advanced analytics that is still largely untapped in the healthcare industry [4]. Healthcare and clinical providers generate billions of provider-patient exchanges, medical histories, insurance claims, care guidelines, research papers, and regulatory reports, with in-depth content and far-reaching implications. The potential business benefits of mining this large pool of unstructured data is enormous, for patients, providers, insurers, pharmaceutical companies, government, and legal entities alike.
However, the analysis of text is not so simple. It relies on very strong and committed human judgment. Linguistic constructs often require interpretation, and values of sentiment often have to be determined based on context. Add to that the complexities encountered in multi-lingual societies where people mix languages and terms, and social media, where all kinds of symbols, abbreviations and short hands are liberally used. Textual data often requires cleansing, dictionary-assisted translation, the right tools and especially the correct algorithms that must be applied. Most importantly, a large amount of learnings must be ploughed back into the process. The models’ effectiveness must be measured, and approaches and techniques must be adjusted accordingly. No software can do this automatically. It requires creative thinking by skilled analysts with relevant analytical experience in both data mining and text analytics.
Conclusion
I have shown three areas of advanced analytics that can be very advantageously applied in the healthcare industry, and which up to now have only been applied at the surface level or in isolated pockets. The biggest gains will be achieved when organisations start embracing the internal and external integration of healthcare data and analytics. Both within and across organisations, the healthcare analytics of the future need to cross boundaries. In healthcare organisations we often find analytical silos – disconnected and isolated analytical groups that do not share data, technology or expertise. There is virtually no analytical integration across the payers, providers and life science organisations. When that changes, we will see an explosion of the value that the application of advanced analytics can bring to organisations in the healthcare industry.
References
1. Steffes, T., Predictive analytics: saving lives and lowering medical bills, published in www.analytics-magazine.org, January/February 2012
2. Norman, G., 2007, “It takes more than wireless to unbind healthcare,” presentation at Healthcare Unbound Conference.
3. Roebuck, M. C., Liberman, J. N., Gemmill-Toyama, M. and Brennan, T.A.,
2011, “Medication Adherence Leads To Lower Health Care Use And Costs Despite Increased Drug Spending,” Health Affairs, January 2011, Vol. 30, pp. 191-199.
4. Grimes, S. and Anderson, T.H.C., Infinite Possibilities of Text Analytics, http://www.textanalyticsnews.com, March 2012.
5. Advanced analytics are a powerful secret weapon against healthcare fraud, published in Managed Healthcare Executive by Andrea Allmon, PHD in August 2005
