Improving Clinical Decision Support with Data Analytics
If you are somehow â€“ even tangentially â€“ connected to the healthcare industry, there is no escaping healthcare reform. Whether one is a vendor, part of a health plan, part of a health system, or is simply a patient, the migration away from fee-for-service to value-based payment models such as global capitation, shared savings, and bundles along with the increasing requirement that patients pay for larger portions of their healthcare (through co-insurance requirements, higher deductibles, and increasing total out-of-pocket requirements) impacts every facet of the healthcare business and oneâ€™s interactions with it. Building on top of the move towards value-based payments and increasing patient responsibility is the effort by many parties in healthcare to operationalize the Triple Aim â€“ increasing quality, decreasing cost, and increasing patient satisfaction or, for the more ambitious, the Quadruple Aim which is the Triple Aim with the addition of vocational satisfaction amongst those in the healthcare industry (especially clinicians).
Operationalizing the Triple or Quadruple Aim to adapt â€“ or, preferably, to thrive â€“ in this new environment requires that organizations embrace many of the tools that other large industries have used to become both more efficient and to meet the needs of their customers. Industries such as manufacturing, supply chain management, and finance have used process improvement strategies such as Six Sigma and LEAN to redesign operations to reduce waste, increase profitability, and increase employee satisfaction. Healthcare is not immune from this; indeed, the push towards data-driven decision making using analytics and clinical decision support tools (CDS) is often the backbone of such transformational activities.
Healthcare analytics encompasses all the activities that result in the analysis of data collected from a variety of sources such administrative or claims data, unstructured or structured EMR data, financial data, supply chain data, and patient reported data such as satisfaction surveys. Often the analysis is concentrated on improving operations of a health system (e.g., financial or supply chain) or for clinical process improvement (e.g., ensuring that all diabetics have received retinal eye exams and A1C screenings). A corporate culture that emphasizes seeking answers to questions in data and a commitment to data-driven decision making is wholly dependent on the existence of enterprise-wide analytics. To implement the methodologies mention above – e.g., Six Sigma, comprehensive analytics systems ought to be available to ensure that the key performance indicators can be tracked and that component processes and sub processes can be measured properly.
Building upon healthcare analytics, clinical decision support or CDS are a set of tools that assist in clinical decision making by linking known data on individual patients and populations to existing overall health knowledge to optimize care and reduce adverse events and errors. A simple example of a CDS process is one that sits on top of a medication module in an EMR. When a provider begins prescribing a medication, the system will alert the provider proactively if a medication that is being prescribed is known to adversely interact with an existing medication the patient is currently taking. Such a configuration is a common albeit rather primitive form on a CDS. Some CDS systems, however, have more advanced features and rather than simply working from existing knowledge sets – e.g., drug interaction databases – they use the entire data set available within the system to infer new trends that may assist clinicians in determining optimal treatment plans or to, at a minimum, provide inferences to assist clinicians in working difficult cases.
Drivers for Advanced Analytics
The healthcare environment is intensely competitive today. Different health systems are working diligently to recruit the best providers to build networks that are attractive to health plans, employers that self-insure, or to use within their health plans – especially Medicare Advantage plans. In addition to network recruitment, healthcare enterprises are under intense pressure to quickly adapt to new payment models that emphasize cost containment, accurate risk stratification, and process improvement. Such payment models such as shared savings, capitation, and episodic bundles radically change the payment methodology that healthcare enterprises are under; fee-for-service does not incentivize such behaviors since, under most circumstances, services rendered result in payment regardless of whether or such services were the most efficient care pathway for a given condition or if a service could have been prevented had better preparation occurred.
The financial pressure caused by payment reform has created strong incentives to establish effective analytics infrastructure. Powerful analytics provide enterprises with information on how much the total cost of care is for patients, how much care patients receive in and out of the healthcare network, what procedures/services were performed, and EMR data – labs, structured preventative data, etc… Outside of the clinical area, enterprises specific financial information and supply chain information is often also available in exhaustive analytics systems. Many analytics systems also have patient survey and satisfaction data.
This data allows enterprises to determine which procedures, for example in an episodic bundle, are the best to take risk on and which procedures the enterprise is not particularly efficient. For example, a given health system may have exemplar performance with total knee replacements; however, the same enteperise may be more expensive than the national and regional averages for balloon angioplasty procedures. Thus, the enterprise might initially enter into episodic risk contracts for total knee replacements that cover all care from the day of surgery to ninety days post-discharge. The same organization, however, may defer entering into such agreements for balloon angioplasties.
In such a scenario, a data-driven health system would have separate teams convene to strategize how to further improve total knee replacements by looking, for example, at how individual providers perform or if there are opportunities within the supply chain to standardize equipment used. With balloon angioplasty procedures, however, different questions may be asked and existing protocols successful elsewhere may be studied to assess how to align the enterpriseâ€™s cost with national and regional averages. Such uses of analytics will allow clinical and administrative stakeholders to study, when necessary, individual cases in detail to use them as examples to build new processes to decrease the likelihood of adverse events and increase overall patient safety. Powerful analytics systems can automate part of the process to quickly provide more accurate data to decision makers. Stakeholders can also use advanced analytics systems to determine which providers deviate greatly for the enterpriseâ€™s average – cases that deviate from the average could be either examples of optimal behavior or opportunities for improvement.
Such uses for analytics are beneficial for the other main driver for advanced analytics – public/governmental involvement. Increasingly, health systems are being required to report their quality information and are liable for reimbursement penalities from CMS due to their performance in certain areas – e.g., for hospital acquired infections, readmissions, falls, etc… The quality improvement necessary to decrease all-cause readmissions or to identify trends that lead to above average hospital acquired infections are dependent on powerful, multi-source analytics that allow an enterprise to examine processes at both the micro and macro level to identify opportunities for improvement; for example, to reduce readmissions, care managers may work with certain subsets of patients on more outreach post-discharge or a remote patient monitoring program may be put in place for patients with congestive heart failure (CHF). These processes and systems can feed data back into the analytics system, and new the clinical decision support system can subsequently proactively inform clinicians when a patient is a good fit for such a program. Increasing the quality of care for patients to either end penalities leavied by CMS or to avoid them altogether also feeds back into the financial driver as, typically, efficient care – e.g., care that reduces unnecesssary care – tends to be lower cost as unnecessary services are no longer rendered. Institutions with fewer penalities and higher publically reported quality are likely to also be more attractive to purchases of healthcare and patients.
Establishing a Competitive Advantage
It is increasingly evident that the changes thus described in healthcare will cause significant challenges for organizations unprepared and unable to readily adapt to new processes and initiatives that lead to better cost and quality outcomes; moreover, it also is likewise evident that organizations that can and do make some adaptations will likely to be greatly sought after by patients, employers, and health plans but will also thrive in such a dynamic, risk-based payment environment.
Building a feedback loop that retrospectively studies cases and allows analysts and clinician leaders to query comprehensive data sets to determine opportunities for improvement or to lower deviations amongst providers is paramount for success in this new environment. The new insights can be monitored for success by quality improvement personnel within the analytics system and, where applicable, new alerts and modifiecations can be added to clinical decision support tools to ensure that new protocols and processes are appropriately used. With financial operations, enterprises can seek out opportunities to standardize tools used to lower costs or further expand the use of group purchasing organizations (GPOs) to lower costs.
Advanced analytics can be used to, for example, refine remote patient monitoring protocols for CHF or chronic obstructive pulmonary disease (COPD) patients to decrease ER usage for symptom exacerbation and, likely more importantly for the patient, increase the likelihood that the patient can remain in their home longer; the result, in addition to lower cost of care and an increased chance of profiting from risk-based contracts, the patient is likely more satisfied and thus apt to recommend the health system to others seeking care. That is extremely important; successful and profitable systems in the new healthcare world are sought after for treatment rather than defensively trying to keep care in-network.
Enterprises that invest in a strong analytics infrastructure are also enabling themselves to be prepared for compounding their successes with analytics and process improvement initiatives as machine learning continues to evolve. Machine learning will enable enterprises to use to the data warehouses that are established for analytics to be used to develop new insights proactively as the advanced algorithms merge the existing data with known scientific data to produce idea treatment regimens for patients. Currently, this is being used with some oncology patients as machine learning tools merge clinical data on the patient with known scientific data to perform actions such as confirming diagnoses and recommending evidence-based treatment plans.
As this technology moves beyond oncology and other select specialties, it will likely be able to predict patients that are likely to be potential readmissions and thus allowing the facility to proactively address barriers to keeping the patient in an ambulatory setting â€“ e.g., establishing a primary care relationship, addressing a social determinant of healthcare, etc.. Machine learning may also be used to predict the best of multiple treatment pathways for patients in a bundle care arrangement by, for example, determining patients that are optimal for receiving a procedure at an ambulatory surgery center in combination with a skilled nursing facility for rehab. There is also the possibility of recommending specific medicines â€“ where numerous are available â€“ for patients with chronic conditions such as major personality disorders or coronary artery disease. Using known scientific research in addition to the patient profile, a successful algorithm can predict optimal medication plans faster than a clinician; thus, the clinician can concentrate on spending more time with the patient, rather than combing through scientific studies.
What Kind of Organization?
That is the question enterprises must ask themselves. If an organization wishes to proactively adapt to the future of healthcare and to be a sought after by purchases and consumers of healthcare, they must prepare themselves by establishing strong clinical and quality improvement leadership. Data cannot solve problems, people do; nevertheless, even the most capable individuals require good tools to succeed. In this changing healthcare environment where leadership must confront increasingly assertive patients, payment reform, and regulatory reporting requirements, a strong analytics infrastructure is necessary to unleash the potential of capable clinical and quality improvement teams. Organizations must embrace the call to big data and the inevitability of machine learning if they are to be successful in an increasingly competitive environment marked by increasing acquisitions, mergers, and more demand from payers and employers that the providers of healthcare take on additional risk for the population under their care.