NICE REPORTS, BUT NOW WHAT…. 12 NEW QUESTIONS POPULATION HEALTH ANALYTICS SHOULD INFORM
Coleridge’s ancient mariner laments, “Water, water everywhere, nor any a drop to drink.”[1] Having spent the last six months exploring the sea of population health analytics platforms, I empathize. Some vendors inspired my rhyme, “Lousy vendors everywhere, away these guys must slink.” In 2016, there were over 100 population health management companies exhibiting at HIMSS.[2] Obviously, they are not all great. However, even among the best solutions, all of whom can be quite useful, there is room to improve. As the mariner might put it, “Data reporting everywhere, but insights and actions don’t yet link.” Specifically, solutions need to better anticipate the types of questions users will ask of them. These questions will be driven by the actions population health managers can realistically take to improve care efficiency and quality.
While definitions vary, I typically describe Population Health Analytics platforms (PHAs) as tools that aggregate and report on claims, EMR, and other data, with the goal of improving care quality and reducing costs. All PHAs present clinical and financial metrics at patient-, provider-, practice-, and population-levels. They have a library of industry standard and proprietary quality metrics and potential gaps-in-care. They embed to varying degrees diagnosis-based predictive models, e.g., HCCs, ACGs®, for risk prediction and reporting normalization. They can all create patient registries, and all have some form of ad hoc reporting. Typically, PHAs can benchmark client organization performance against national averages. Significant variation exists among PHA solutions within this functionality set. Some PHAs are better at certain tasks. (See Chilmark’s 2016 Analytics for Population Health Management Market Trends Report for an excellent overview of the leading players.[3]) I contend that all PHAs platforms have a common shortfall. They are all organized around the data sources they report on, rather than around the questions that users want the data to answer. Even so, with a little clicking and some occasional exports to EXCEL, all the PHAs can at least somewhat answer the following questions:
- Which providers and patients are cost outliers?
- Which patients may benefit from more attention?
- Which quality metrics require more attention?
- How do broad categories of utilization and spending differ from benchmarks?
- What types of care are leaking to third party providers?
A fine start, but there are many more questions the data can answer. The list below is but a small sample to spur thinking.
1. How can I better manage high-prevalence, high-cost diseases? (Condition-specific analytics) Most PHAs support some disease-specific quality metrics, e.g., are diabetics on Statins, or are CHF patients on anti-hypertensives. Many report disease-specific outcomes as well, e.g., HgBA1C, cardiac ejection fraction, and BMI. However, for a variety of reasons, most do not go much deeper. For example, few analytic solutions allow for easy analysis by disease stage, e.g., CKD 3 vs CKD 4, or Class 3 vs Class 4 CHF. Similarly, understanding the impact of different treatments is challenging. For example, most PHAs are not well equipped to compare outcomes of diabetics on oral-meds only vs. insulin-only vs. oral-meds with insulin. As new high-cost treatments for common diseases come to market, e.g., PCSK9 inhibitors for hyperlipidemia, Entresto® for CHF, protease inhibitors for HEP-C, disease-specific understanding will become even more important. Further, certain high-cost conditions, e.g, pregnancy and ESRD, have very specific analytic needs that are often all but ignored by current PHAs.
2. How do I assess the total cost of care for routine services? (More bundles) Many PHAs have already implemented some care bundle analytics. These allow population managers to evaluate whole episodes of care, e.g., a joint replacement, a CABG. Work on this should continue. In addition, the types of episodes should be expanded. For example, rather than simply figuring out who provides the cheapest mammograms, a bundle could incorporate all the downstream breast imaging associated with the initial study. Such a payment mechanism could reduce unwanted variations in treatment, and help funnel volume to better performing providers.[4] Other opportunities include colonoscopy and certain cancers. Even if these new bundles are not as well defined as joint replacement, they are likely still useful.
3. What am I doing today that does not need to be done? (Unwanted care variation) Quality metrics within existing PHAs already help reduce variations from evidence-based guidelines. However, the clinical metrics embedded in today’s solutions, rarely totaling over a hundred, just scratch the surface. Consider that there are 271 measures in the MIPS Quality Payment Program alone.[5] Many of these are targeted specifically to standardize treatments, e.g., reduce use of unneeded antibiotics, drive appropriate screenings, and initiate of protective medication regimens. ABIM’s Choosing Wisely® initiative convened the leading US medical societies to develop lists of unnecessary treatments.[6] There are now over 400 such recommendations. Even with these efforts, there is more to be done. For example, Geisinger Health System measures 40 different clinical activities for their CABGs alone.[7] Doing so allows them to provide warranties on the procedures. As the number and complexity of metrics proliferate, it would be very helpful if PHAs not just report them, but algorithmic-ally prioritize them in terms of patient care impact and cost.
4. Is my care being delivered in the right setting? (Site of service analytics) Shifting care to lower-cost settings is a powerful lever for population health management. For example, many procedures can safely be done in a variety of locations, e.g., a physician office, an ambulatory surgery center (ASC), or a hospital outpatient department. The price differences between them are dramatic. One study found that converting an ASC to a hospital outpatient department can increase reimbursement rates by 81%.[8] At the same time, other studies have shown provider ownership of ASCs can increase utilization.[9] Similar dynamics can be seen in provider-ownership of high-end imaging.[10] In another example of cost control via site-of-service migration, consider minor injuries. These can often be treated via telehealth visits, physician office visits, at urgent care centers, or in the ED. Each of these locations can have markedly different unit costs. In parallel, each of these locations see different levels excess utilization, which can markedly change overall economics. For example, at ~$50 televisits might be very cheap, at least compared to a $500+ ED visit. However, there is meaningful potential for overuse.[11] To unravel all this unit cost and utilization interplay, PHAs need better site-of-service analytics.
5. How can I reduce my drug spend beyond generic substitution? (Medication selection and pharmacy channel analytics) Generic substitution has been a major contributor to lower drug spending.[12] However, since 2016, dollar sales of branded small molecules going off patent have been declining.[13] Accordingly, there will be fewer near-term additional savings opportunities from increased generic use. At the same time, pharmaceuticals as a percent of total spending has been steadily rising. Thus, more sophisticated medication cost management programs are needed. These will require more sophisticated analytics. Low hanging fruit include therapeutic substitution analytics, dispensing pharmacy analytics, and retail/mail analytics. More advanced analysis includes effectiveness of managed-access programs (e.g., step therapy, quality limits, prior auth), adherence rates of different medications, and outcomes of different pharmaceutical strategies. Specialty medication analytics, especially for oncology, offer even greater opportunity to reduce costs and standardize treatment.
6. How can post-acute care be optimized? (Post-acute & transport analytics) Post-acute services are a major component of overall healthcare spending, especially for seniors. Long-term care hospitals, inpatient rehab facilities, and skilled nursing facilities account for a combined 18% of Medicare spending.[14] The post-acute care experience impacts overall health costs and patient satisfaction.[15,16] Within post-acute providers the variation is tremendous. Remedy Partners® found “top quartile [SNFs] not only spent almost half per SNF episode compared to the bottom quartile, but readmission rates were also halved.”[17] Better coordinating post-acute care creates significant cost reduction opportunities. Huckfeldt et al. found that patients covered by private Medicare payers received less intense post-acute services and experienced better outcomes.[18] Relatedly, variations in medical transport utilization are also large and thus potentially manageable.[19] PHAs that shed light on post-acute quality and price discrepancies can create some immediate wins.
7. Which patients are intrinsically more likely to succeed? (Multi-faceted patient risk adjustment) Most PHAs can risk adjust patients using their underlying disease burden. For example, if a patient has COPD and breast cancer, they are likely to cost much more than a patient with only hypercholesterolemia. This risk adjustment is critical for identifying which providers are better or worse at managing population costs. It would be unfair to expect a provider with many very sick patients to have the same average spending as those with healthier ones. However, the underlying disease burden of a patient is only one predictor of utilization and outcomes. There are many determinants of health, including genomics, socioeconomics, and access to care. While the current science may not yet support genomic-based risk prediction, data systems already often capture insurance coverage, family income, English proficiency, zip code, and level of education. Enriching the predictive models with this data would likely increase both accuracy and clinician-acceptance.
8. Are my patients getting the behavioral healthcare they need? (Behavioral health analytics) Behavioral health conditions are often poorly diagnosed and hence poorly managed.[20] Further, there are significant variations in the delivery of behavioral healthcare within primary care.[21] This is unfortunate for at least three reasons. Most importantly, untreated illness reduces the patient’s quality of life. Secondly, failing to manage the illness can result in costly acute exacerbations. Finally, the presence behavioral health conditions impact the cost of treating “organic” diseases (e.g., diabetes, hypertension, CKD). Milliman found that chronically ill patients with a co-morbid mental health/substance abuse disorder cost 2-3 times more than a similar patient without a behavioral condition.[22] Analytics that can help detect under-diagnosis or under-management of behavioral conditions, and appropriately risk adjust expected total spending taking this into account, will be valuable to population managers.
9. Is the care coordinated between various providers? (Referral analytics) Whether a health system provides most of their patients’ care themselves or refers out extensively, it is crucial to understand referral patterns and the performance of patient care teams. Inappropriate care fragmentation has significant impact on quality and costs. Hence the need for better referral analytics. Referral analytics is a broad term. For primary care, it includes level of specialist referrals by patient type, appropriateness of referrals as defined by system-specific guidelines, patient follow through in getting the care, concentration of referrals to higher performing specialists, and degree of co-management with specialists. For the specialists, referral analytics could include timeliness from referral to appointment, level of sharing of management with primary care, and direct referrals to other specialists. Finally, such analytics can help identify de facto provider networks who appear to collectively have better or worse outcomes than their peers.
10. Which specialists are providing better care than others? (Specialist performance analytics) Certain specialties, especially high-volume surgical ones, lend themselves nicely to performance analytics. Metrics are clear, e.g., infection rates, readmissions rates, mortality, episode cost, discharge-to-home rates, and length-of-stay. The volume is large enough that the data is statistically significant. An ophthalmologist doing 15 or more cataract surgeries a week generates a lot of data. Unfortunately, other providers, most notably oncologists, lack high volumes of similarly ill patients. Even among common tumors there is tremendous variation in stage, tumor cell pathophysiology, and relevant underlying patient genetics. Hence, it is hard to perform analysis on their outcomes. The sample size is not large enough. Other specialties, including rheumatology and neurosurgery, face similar challenges. Accordingly, PHAs must not just rely on a health system’s own data for specialist evaluation. Any one health system’s patients may only be a small fraction of a specialist's volume. Rather, PHAs must aggregate multiple data sets to achieve a fuller picture of clinical performance.
11. How are my patients feeling about their care? (Patient-generated outcomes measures) Healthcare, be it evaluated at the individual or population level, is a service business. All quality and efficiency improvement initiatives must ultimately be viewed from the patient’s perspective. Standardized, analyzable data on patient reported outcomes and satisfaction is crucial to program design and evaluation. Of course, for many risk-based contracts, satisfaction itself is an outcome measure that directly impacts compensation. This area is woefully ignored by many otherwise high quality PHAs.
12. How can I save my patients money? (Financial burden analytics) Most PHAs focus their financial reporting on “plan paid” amounts. These are the dollars insurers pay for any service. The reason for this is simple—most shared-savings or other risk-based contracts are tied to these dollars. Hence, the patient’s share of the health cost is often overlooked. This is problematic. Firstly, out-of-pockets costs can directly impact patient satisfaction. Secondly, cost control is an issue patients specifically want their providers to help them with. A survey by Agate Consulting found that 39% of patients want their providers to improve their ability to help patients control out-of-pocket costs.[23] This ranked appreciably higher than improving care quality, better coordinating care, and better using IT. Finally, out-of-pocket cost exposure can reduce patient adherence to medication use [24], preventative care [25], or other elements of the care plan. Analytic tools that identify compressible high out-of-pockets costs can help alleviate this burden.
No single vendor could possibly implement all the items above, even if they had nothing else on their product road map. In fact, some of the points above are sufficiently broad to warrent stand-alone solutions. In the meantime, PHA vendors already have multiple albatrosses around their necks. They need to develop better data extraction and cleaning methods, continually meet ever-changing regulatory needs (especially MACRA), improve data security, support emerging payment mechanisms (e.g., bundles), enable bi-directional EMR interoperability, and improve general usability. With that said, I strongly encourage PHA vendors to constantly consider “what questions are my customers asking themselves,” and “what questions should they be asking themselves?” Only by starting with the potential interventions in mind can population health analytics effectively enable health delivery transformation. Population health analytics is very long journey—something the ancient mariner would fully appreciate.
NOTES:
Coleridge ST. The Rime of the Ancient Mariner, 1834. This line is often misquoted as “Water, water everywhere, but not a drop to drink.”
Wardell N. Population Health Management: Understanding the Implications of all the Talk at HIMSS16. TripleTree Holdings, LLC.
2016 Analytics for Population Health Management Market Trends Report. Chilmark Research, Inc., purchasable from https://tinyurl.com/y6w9afe9.
Hughs DR et. al. An Empirical Framework for Breast Screening Bundled Payments. J Am Coll Radiol. 2017 Jan;14(1):17-23.e1.
CMS’s Quality Payment Program website at https://qpp.cms.gov/mips/quality-measures accessed on 6/14/2017.
ABIM Foundation web site at http://abimfoundation.org/what-we-do/choosing-wisely.
Casale AS et. al. ProvenCareSM. A provider-driven pay-for-performance program for acute episodic cardiac surgical care. Ann Surg 2007;246: 613-623.
Nantz J. HOPD or ASC: 5 questions for hospitals to consider. Becker’s Hospital Review. Mar 10, 2015.
Hollingsworth JM et. al. Physician-Ownership of Ambulatory Surgery Centers Linked To Higher Volume of Surgeries. Health Aff April 2010 vol. 29 no. 4 683-689.
CHRT. Physician Ownership in Hospitals and Outpatient Facilities. July 2013.
RAND Corporation. Direct-to-Consumer Telehealth Prompts New Use of Medical Services; Not Likely to Decrease Health Spending. Mar 6, 2017 press release.
Medical Cost Trends: Behind the Numbers 2018. PwC Consulting.
Ibid.
CMS’s Geographic Variation in Standardized Medicare Spending dashboard at https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-Reports/Dashboard/GeoVar-State/GeoVar_State.html.
Variation in Health Care Spending, Target Decision Making, Not Geography. Institute of Medicine presentation.
Trend watch. The Role of Post-Acute Care in New Care Delivery Models. American Hospital Association. Dec 2015.
Olexa C. Narrow SNF Networks and Increased Collaboration: The Remedy Approach. Remedy Partners blog. Feb 16, 2016.
Huckfeldt PJ et. al. Less Intense Postacute Care, Better Outcomes for Enrollees in Medicare Advantage than Those in Fee-For-Service. Health Aff January 2017 vol. 36 no. 1 91-100.
Hanchate AD et. al. Geographic Variation in Use of Ambulance Transport to the Emergency Department. Ann Emerg Med. 2017 May 27. pii: S0196-0644(17)30325-6
Basco MR et. al. Under diagnosing and Over diagnosing Psychiatric Comorbidities. Psychiatric Times. Oct 1, 2008.
Kwan B et. al. The State of Evidence for Integrated Behavioral Health in Primary Care. Chapter 5 of Integrated Behavioral Health in Primary Care ISBN 978-1-4614-6889-9.
Melek SP et. al. Economic Impact of Medical-Behavioral Healthcare. Milliman. Apr 2014.
Agatstein KA. What’s in the is MACRA thing for me. Agate Consulting white paper. Jan 2017.
Eaddy MT et. al. How Patient Cost-Sharing Trends Affect Adherence and Outcomes. P T. 2012 Jan; 37(1): 45–55.
Dolan R. From The Archives: Deductibles and Out-Of-Pocket Costs. Health Affairs blog, Sep 29, 2015.