A Geriatrician’s Guide to Accountable Care Implementation: Thickets and Pathways

Abstract

The focus on value and accountable care opens unprecedented opportunities to create effective care systems for older and/or high-need, high-cost patients. The journey to accountable care is far from linear, however. Multimorbidity and the social complexity of these patients resist easy off-the-shelf interventions, particularly given the entrenched fragmentation across primary and specialty care, post-acute and long-term care, palliative care, and community-based services. Each delivery system must find its own way across a landscape strewn with false starts, failures, and avoidable suffering. This chapter will review a broad array of innovations with particular attention to program development and adaptation, as well as the levers of organizational change. Readmissions are one such lever. Readmission reviews provide line of sight into the interstitial spaces of our health delivery systems. Readmission reduction entails addressing multiple services across multiple sites of care, e.g., medication management, advance care planning, and palliative care. Similarly, new partnerships in post-acute care and community care promise to reduce high-cost facility utilization. Predictive analytics tools help match high-touch resources to patients with remediable needs. Your ability to develop and sustain geriatric programs will depend upon your ability to obtain and present credible data on clinical and financial performance. Challenges remain, but the focus on value has begun to align the transactional logic of operating margins and the mission-driven logic of healing relationships. We should take advantage of this overlap wherever we can.

Appendix on Measurement: Necessary, Potent, and Imperfect

If your delivery system has begun taking risk for a defined population, then your information systems presumably include a data warehouse supplied by claims, (facility, professional, laboratory, radiology, pharmacy), authorizations, and patient experience data. You may be missing some of these elements; you may also have additional elements, e.g., clinical data from electronic health records and care coordination systems. At a minimum, your business intelligence capacities will yield the performance measures shown in Table 6.2. While you may be most interested in data related to older and/or high-need, high-cost populations, these need to be understood in the context of the delivery system performance as a whole, since that is the perspective the system’s decision-makers must take.

Table 6.2 Typical claims-based performance measures for various providers

Monitoring these basic measures will reveal variation across time and across providers (physicians, acute and post-acute settings) and thus serve as a starting point for your questions. Your business intelligence capacities should assist your search for answers by allowing you to drill down to the individual provider, e.g., clinic or hospital, and to trend performance over time. Additionally, you will need to filter measures by line of business and product and to segment if possible by risk tier, diagnoses, and service lines. While it is important to identify, understand, and address negative outlier performance, it may be more valuable to identify, understand, and replicate superior performance.

In addition to these claims data, most ACOs have access to measures of patient experience. Note that utilization may be more illuminating than cost when there are significant differences in unit costs and/or missing data on costs. When possible, you should take advantage of authorization data, which have shorter lag times than claims.

Measurement in Older and High-Risk Populations

• Claims-based quality measures include the Medicare Stars specified by CMS, e.g., screening for colon cancer and renal insufficiency and management of diabetes and osteoporosis.

• Of particular interest for management of older populations are:

  • Annual wellness visits

  • Primary care claims for care coordination and care transitions

  • Lag times from hospital discharge to outpatient follow-up and nursing facility discharge to outpatient follow-up

• Claims will enable at least bare-bones monitoring of specialized programs, e.g., the patient enrollment, visits, and facility utilization of a home-based medical care program. There is interest in development of standardized quality metrics for such programs, but these as yet do not exist [214].

• Claims will enable identification of “hidden” high-risk populations such as patients in domiciliary settings (assisted living, board-and-care homes) or custodial nursing facilities or on dialysis.

• To monitor the channeling of patients to preferred post-acute home health agencies and nursing facilities, I suggest using the Herfindahl-Hirschman Index, which is the standard formula for market concentration. Your goal is to concentrate patients as much as possible with preferred providers.

You may find useful variation in monitoring claims for advance care planning, although those claim numbers are not a measure of the quality of advance care planning conversations. Similarly, you are likely to find variation in the volumes of inpatient and outpatient palliative care consultations. If you access dates of death from state or federal agencies and match those data with your claims, you will be able to create robust end-of-life quality and utilization measures reflecting the use of primary care, specialty care, emergency department, hospital, and chemotherapy in the patient’s final weeks and months.

Risk Adjustment and Socioeconomic Determinants of Health

Risk adjustment is critical for comparing performance across providers, but it continues to be challenging. The Hierarchical Condition Category (HCC) model employed by CMS does not adequately account for the resource use of complex older patients [215]. There is now intense interest in using socioeconomic determinants of health as adjustments in risk estimates, quality measures, and payment, but methodologies are not yet in widespread use [216].

Socioeconomic factors also have potential clinical utility in patient management and program development, particularly for high-need, high-cost populations. The SCAN Foundation report discussed earlier outlined four essential attributes of delivery systems caring for adults with complex care needs. The National Quality Forum has since reported on data and measurement systems that are being used to support progress toward the four essential attributes [217]. One of the promising examples described is the Protocol for Responding to and Assessing Patients’ Assets, Risks, and Experiences (PRAPARE). This 22-question tool captures biopsychosocial determinants of health, including race and ethnicity, housing status, neighborhood, social integration, and social support. It is freely available for incorporation into electronic health records.

Characteristics of Good Measures

The simplest way to evaluate a given measure is to ask, How accurately does this measure reflect meaningful variation in a process or outcome? Criteria for measure selection typically used by CMS, the National Quality Forum, and the Institute of Medicine [218] include the following:

• Impact: importance for health status and/or cost, i.e., does the measure really matter?

• Improvability: existence of gap between current practice and best practice, as well as evidence that the gap can be closed (whether the measure is actionable).

• Feasibility, including data availability and the burden of data collection.

• Scientific soundness and methodological rigor of the measure, including validity (credibility, or how well the measure captures the process or outcome it is intended to assess) and reliability (consistency, or whether the measure produces similar results under consistent conditions).

• Understandability of the measure, e.g., obvious specifications versus “black box.”

• Timeliness: the turn-around time and frequency of measurement (e.g., monthly versus annually).

Measurement criteria become more stringent along a continuum from low stakes, e.g., quality improvement feedback, to high stakes, e.g., payment for performance.

• For rapid-cycle quality improvement purposes, low-rigor measures and nonrandom small samples can suffice for adequate insight, as long as the stakes as perceived by providers are low.

• Imprecise or “noisy” but directionally accurate data may suffice when an organization is giving performance feedback to a provider but not sharing the data more widely.

• Clinicians within a setting, practice, or specialty may also appreciate getting variation data regarding a process when the “correct number” or Goldilocks optimum is unknown. Such data can trigger a useful conversation within the group members about why they vary in their decision-making.

• Sharing data openly throughout a specialty can have significant positive impact among physicians, who are typically quite competitive, but doing so also increases the demand for rigor. Sharing data on specialists with primary care physicians can influence referrals, increasing the stakes even higher. Paying incentives for performance multiplies the demand for rigor.

Measurement Challenges

Increasing recognition of the burdens of measurement has led to calls for reduction and harmonization of measure sets. In addition, CMS has recently emphasized the unintended consequences of performance feedback and incentives, including worsening quality in unmeasured areas (“teaching to the test”), providing overtreatment or unnecessary care, gaming of the data, avoiding high-risk or challenging patients, and worsening disparities in care [219].

Many clinical domains lack meaningful measures that can be derived from existing data sources. Before giving up on a domain, delivery systems should consider the use of manual sampling strategies. Guidance for efficient and reliable sampling is readily available [220]. It remains true that some important domains defy direct measurement. We have no measures addressing the act of diagnosis, for example, yet that act is central to the practice of medicine [100]. We often find ourselves looking for substitute measures that may be distant from the area of interest, rather like the story of the drunk looking under a streetlight for the keys that he dropped in a distant dark alley. That said, good estimates of key processes are invaluable in quality improvement and program development, and we now have excellent guides for improving the quality of our estimates [221].

Attribution of responsibility is often a nontrivial task. Responsibility for an outcome such as readmission, for example, is shared across settings. Additionally, although appropriate attribution of patient to primary care provider is largely straightforward in managed care, it is less precise in CMS ACOs and in managed care populations with high turnover. An obvious question, most pertinent in the pay-for-performance context, is whether the provider has a meaningful amount of control over what is being measured.

The unit of analysis often dictates what measures may be available. The unit of interest may be a county population, a health plan’s enrollees, a medical group’s enrollees, a practice, or a provider. As the analysis becomes more fine-grained, e.g., down to the physician level, small denominators may render a measure unusable because of unreliability (more noise than signal) [222]. The most common approach to the small denominator problem is to exclude providers below a certain cutoff, e.g., 30 patients in the denominator [223]. Bayesian reliability adjustment (smoothing) is sometimes used—and would arguably be fairer to all—but is less intuitive.

Measures that are not robust enough for public reporting or pay for performance may nevertheless be invaluable for the other uses noted above. Also, whereas “noise” may overwhelm “signal” in a one-time use of a measure with small denominators, the signal-to-noise ratio improves with repeat use over time.

Finally, you should not be surprised or overly dismayed to find that you sometimes have difficulty getting reliable data from your information systems. You are not alone. No one has a single electronic health record that unifies all sites of care, and critical data may reside in any of dozens of other applications. Your delivery system encompasses multiple separate organizations contracted to care for your patients. Data flows across these applications and organizations are immensely challenging [224], and few organizations of any kind have invested adequate resources in data management [225]. Decision-makers may remedy this deficit as pressures to manage performance—and the need for measurement insights—continue to increase [156].