We conducted a series of analyses using episodes of care generated by two commercial episode groupers, Symmetry ETGs and Thomson MEGs to explore issues related to constructing and using episodes of care the purposes of measurement and aligning incentives to deliver high quality care. The study population for this work consisted of FFS beneficiaries who were continuously enrolled in Medicare from 2004-2006 and whose 2005 primary residence was in Florida , Oregon or Texas .
The episode groupers utilize the primary diagnosis on claim line items to create and place the line items into an episode. Only certain types of claims can start an episode such as an inpatient admission or an Evaluation and Management (E&M visit). Chronic condition episodes are predetermined to be of one-year duration. For other episodes that do not represent a chronic condition, they are defined by having a “clean period” during which no claims for that condition can appear before a new episode of the same type can start.
Our analyses focused on individuals with a diagnosis of one of nine clinical conditions that were purposively selected to illustrate various issues, such as discrete time-limited events that might entail fewer providers and/or settings, chronic conditions of long duration that might involve management by a broad collection of providers, and complex events that would likely entail care provided across an array of settings of care. The nine conditions were:
- Acute myocardial infarction
- Bacterial pneumonia
- Breast cancer
- Cerebrovascular disease
- Chronic obstructive pulmonary disease
- Congestive heart failure
- Hip fracture
- Low back pain
For each individual with one of the nine clinical conditions, we categorized all of their episodes that were constructed by the grouper tools into those that were “related” to or “unrelated” to the condition for which they were selected. It should be noted that the results that we observed are, in part, related to how the episode grouper tools define what claims get assigned to an episode (i.e., the underlying grouper logic used to construct an episode) as well as variations in coding practices among providers in what diagnosis they code as primary versus secondary (and the completeness of this coding). The primary diagnosis drives the start of an episode of a particular type.
The key findings from our analyses are:
- Medicare Beneficiaries Have a Large Number of Different types of Episodes per Year: Beneficiaries with the nine conditions experienced an average of 10 episodes of any kind during the measurement year, most of which were not related to nine conditions of focus in this study. Many of the unrelated episodes were common among a large proportion of beneficiaries across the nine study conditions, such as hypertension, congestive heart failure, and fungal skin infections. The large number of episodes per beneficiary raises questions about the degree to which care for a particular beneficiary should be examine holistically, or alternatively split into small units of analysis. It is unclear whether physicians and other providers would view a beneficiary’s multiple episodes as defined in this study as distinct issues to be managed separately or as related issues to be managed jointly. If providers view certain episodes as related issues that should be managed jointly, (e.g. episodes of ischemic heart disease, hypertension, and hyperlipidemia), then it may be appropriate to expand episode definitions to group related conditions for some applications. One possibility would be to create bundles of episodes that commonly co-occur and are jointly managed. An issue that needs to be considered but which was not addressed in our analyses is whether the same provider would be attributed primary responsibility for multiple different episodes.
- Standardized Payments per Episode Varied Widely Across and Within the Nine Conditions. An inverse relationship was observed between standardized episode payments and the coefficient of variation, a measure that identifies the amount of variation in payments between episodes related to the same condition. Thus, there is wide variation in what is happening to patients within episodes of the same type that were constructed using commercially-available software grouper tools, suggesting a fair amount of heterogeneity in care practices and/or types of patients being treated. We used fairly broad groupings of patients based on the nine conditions. Heterogeneity might be reduced if subgroups of patients were created within a given condition. For example, instead of grouping together all diabetics, one could separate diabetics into categories based on the degree of advancement of their disease and other existing comorbidities to address differences in the management of these individuals. The amount of variation observed in the analyses suggests a need to understand the sources of variation in standardized payments, and which sources need to be accounted for in the episode construction or patient group creation versus sources of variation that could be reduced through the application of episodes for performance measurement or financial incentives.
Beneficiaries who experienced a greater total number of episodes (both related and unrelated to conditions of focus) had higher average standardized payments per episode and more providers involved in the delivery of care for each episode related to the conditions of focus. This finding highlights the need to consider not only risk-adjusting for the severity of the specific condition of focus, but also the other conditions experienced by the beneficiary.
- The Care Trajectory and Number of Settings Involved varies by Condition and within Episode Types. Across the nine conditions, there was no standard care pattern of the types of providers and settings involved for the related episodes; some conditions were more heavily focused on care delivered in an ambulatory setting, while others involved care delivered in ambulatory, hospital, and post-acute care settings. Even for patients with the same condition, there was substantial variation in the types of settings involved, and it is unclear how this variation in care trajectories would be affected if episode constructions within the same condition were less heterogeneous (i.e., creating more homogeneous subgroups of patients within an episode category). Often care cuts across three settings of care for any given condition and almost 60 percent of hip facture episodes involved more than four settings or provider types, highlighting the importance of care coordination among providers in different settings. During a single episode of care for a particular condition, the care provided was often dispersed among multiple specialists; however, for the nine conditions reviewed in this study, most involved a median of one primary care physician (PCP). These PCPs could potentially provide a foundation for coordinating the care for a beneficiary, if the PCP is also managing care for other episode types a beneficiary may experience. This study did not use a cross-condition approach to examine whether there were multiple PCPs involved in managing a beneficiary’s care across episode types; future work should explore whether there are multiple different PCPs involved in managing care across the entire set of episodes for any given Medicare beneficiary to ascertain whether a single PCP exists to coordinate care.
- Different methods for assigning responsibility for an episode (i.e., attribution) yielded different results. A significant fraction of episodes could be assigned to a provider for most of the attribution rules we tested. Variation was observed in the proportion of the episodes that could be assigned depending on the rule and the type of condition; some conditions are addressed primarily in an ambulatory setting, so facility-based attribution rules led to the assignment of a smaller share of episodes of these types. Depending on the condition, we observed that multiple providers delivered services in most episodes and that some providers represent only a small fraction of total episode payments. Therefore, it is often difficult to determine which provider or setting of care may have had the most responsibility for managing the care and resources within the episode. For example, for episodes where the majority of episode costs are facility costs, which physicians should be held accountable if one were to use a single attribution model? Should it be the physician who managed the patient in the facility or the physician who managed the physician prior to the admission or both? Further, should the facility also be accountable for the episode costs? While most methods of attribution rely on determining which physician may have had the most responsibility, some episodes were comprised primarily of facility costs and therefore it may be important to consider attribution to facilities or multiple attribution to providers and facilities. Given variation in the composition of provider types and settings and the extent of involvement of various providers in the management of episodes for different conditions, attribution rules may need to be tailored to the type of episode to ensure that the assignment aligns with provider roles and responsibilities in managing an episode. A single attribution approach for all types of episodes may not be appropriate. Attribution rules may need to vary depending on the manner in which the information is used and other policy considerations. For performance measurement, multiple attribution could serve to encourage joint responsibility and improvement among all. For resource utilization, the ambulatory physician who could prevent a hospital admission may be appropriate, while for bundling of payment, the entity or entities most able to manage the bulk of the dollars may be a more important consideration.
- State-level Variation Exists and Care within Episodes Cuts Across State Lines: There was variation across the three states in the average number of episodes per beneficiary, both overall and for the subset of beneficiaries with each of the nine conditions, the average standardized payments per episode, the involvement of different post-acute care providers, and the percent of episodes for which beneficiaries received care outside of their state of residence. The mean number of total episodes of all types per beneficiary varied widely among the three states in our analysis, averaging 6.1 episodes per beneficiary in Oregon , 6.9 in Texas and 8.0 in Florida . Average 2005 per-capita payments were highest in Florida and Texas and substantially lower in Oregon . The average standardized payment per episode for the episodes related to the nine conditions varied in a consistent pattern across the nine conditions, with Oregon showing consistently lower average per-episode payments as compared to Florida or Texas . The reasons behind the observed geographic variations in per episode payments and frequency of episodes are unclear and likely reflect several sources of variation, including variations in the care management practices of providers, differences in the availability of and types of providers across health markets, and/or differences in the underlying health status of Medicare beneficiaries in the three states. For example, inpatient rehabilitation facility (IRF) care was more common for episodes in Texas , where these types of facilities are relatively numerous, use of SNFs was more common in Oregon and Florida . A better understanding of the sources of variation could inform the future development of episode-based approaches to quality measurement and payment.