In facilities participating in a demonstration project to reduce PAHs, we identified age, dual eligibility, cognitive function, CHESS, and history of certain comorbidities were associated with the risk of all-cause transfer, including PAHs. These characteristics generally reflect limitations to residents’ functional status and a history of debilitating comorbidities. However, some characteristics were inconsistently associated with the outcomes. Resident ADL score, history of Alzheimer’s Disease/dementia, CMS Star Rating, facility bed count, and participation in the clinical + payment model was associated with the risk of all-cause transfer, but not with the risk of PAHs. Although variation in the risk of transfer was explained at both the resident- and facility-level, resident characteristics explained a greater percentage. These results add to the existing knowledge regarding how to reduce transfers, specifically PAHs, among nursing home residents. Furthermore, risk stratification of nursing home residents by specific characteristics may be necessary for consideration in planning and evaluation of future efforts to reduce PAHs.
The oldest residents (≥ 95 years), the most cognitively impaired (CFS 4), and dual eligible residents had the lowest adjusted risk of both all-cause transfers and PAHs. The lower risk of hospitalization for older residents is equivocal in other studies and may be dependent on how well functional status and comorbidities are measured and controlled for.28 Lower risk in the oldest age group may be attributable to informative censoring, wherein residents in this age group remaining eligible for OPTIMISTIC are overall healthier than those who lose eligibility due to hospice enrollment. To avoid this, some studies exclude observations within 12 months of residents’ death.11 However, due to concerns about sample size reductions and the importance of including events that will enhance our understanding of PAHs, this approach was not feasible. Alternatively, resident characteristics may be confounded by completion of advance care planning/advance directives wherein transfer is inconsistent with their goals of care. Residents in clinical + payment facilities had consistent access to structured Advance Care Planning through the OPTIMISTIC nurses and all facilities had access to tailored educational materials related to palliative care and incorporating goals of care into treatment plans. This may also explain the lower risk of transfer among cognitively impaired residents, and is consistent with other studies of residents with advanced illness and limited life expectancy.29 Although residents with advanced age, cognitive impairment, and/or comorbidities may have clinical features consistent with hospital-level care, decisions to transfer are complex and mitigated by multiple factors that are not captured by the MDS.30–33
In contrast to lower risk of transfer for the oldest residents, residents in the youngest age category—those under 64 years old and thus qualifying for Medicare based on disability—had an increased risk of all-cause transfers, but not PAHs. This finding is consistent with another study where younger age also had functional impairment.34 Furthermore, the lack of an association with PAHs may indicate that hospitalizations among younger residents tend not to be considered potentially avoidable. Moreover, risk of PAHs may not be uniform across all age groups and warrants further investigation.
We observed an increased risk of transfer with higher CHESS scores, which may seem counter-intuitive. One potential explanation is that the CHESS Scale reflects recent health instability, some of which may appear treatable.35 Although CHESS Scale is a predictor of mortality, it may also be sensitive to acute resident changes resulting in transfers, including acute mental status change, dehydration, and pressure ulcers.21 Furthermore, there may be multicollinearity with CHESS Scale and other resident characteristics included in our model, including cognitive impairment. However, variance inflation factors from our models were less than 3 for all variables included.
Facility characteristics including bed size, CMS quality ranking, and presence of the clinical + payment model was associated with reduced all-cause transfers, the associations were not statistically significant with PAHs. Nationally, the overall risk of hospital transfers among nursing home residents has decreased since 2011, coinciding with multiple initiatives to reduce hospitalizations.29 Previous analyses from the OPTIMISTIC project, which has similar components as the INTERACT Program, has demonstrated benefits to improving the management of acute changes in residents’ condition and observed reductions in transfers.13,14,17,18,36,37 Although we do not observe a decrease in transfer risk during our study period, residents in OPTIMISTIC clinical + payment model facilities maintained a consistently lower risk of all-cause transfer than payment only facilities, which increased slightly over time. The OPTIMISTIC clinical model was first introduced in 2012. Thus, nearly half of the facilities represented in our analyses were focused on reducing transfers, specifically PAHs, using specially trained nurses for four years prior to the release of the payment reimbursement codes. The payment component was introduced in the fourth quarter of 2016 and thus the payment only facilities may have had a more heterogeneous experience reducing PAHs.36 Furthermore, the payment only facilities lacked on-site nurses trained to detect and treat conditions to reduce PAHs, instead relevant staff were trained on the use of billing codes for the six conditions.
The absence of an association between PAHs and the clinical + payment model is notable. This may be attributable to limitations of the identification of PAHs from claims data. First, we cannot exclude the possibility that coding practices designed to maximize hospital reimbursement may affect whether a hospitalization was subsequently determined to be a PAH; it is known that administrative claims data provides only a partial picture of the actual clinical experience.38 Furthermore, claims data do not provide direct insights into the primary reason for a transfer—this may be different than what was determined as the final principal diagnosis code for the hospital stay. Avoidability of transfers is difficult to assess, even by nursing staff within the facility, adding to the challenge of making a post-discharge determination using claims-based algorithms.17,39,40 Finally, claims diagnoses reflect the full clinical stay which may include infections or other events that were not present on admission. Nonetheless, our findings that facilities with the clinical + payment model had lower rates of transfers highlights the importance of increasing capacity of facilities to treat patients in place with a proven clinical model such as OPTIMISTIC, and to support practice changes in addition to payment reform.
Although facility characteristics explained a substantial portion of the variation in transfer risk, we observed few with statistical associations. Among them, lower CMS Star Ratings were associated with increased transfers. Notably, initial recruitment of new facilities in Phase 2 of the OPTIMISTIC demonstration project required a minimum 3-star rating. However, facility rating fluctuates quarterly and maintenance was not a requirement, therefore we observed variation within participating facilities over time. It is possible that a decreasing Star Rating over time is indicative of decreasing overall quality of care.
This study has some limitations. First, we recognize this is a non-randomized intervention in 40 facilities within a single state. We do not have the benefit of randomization which could introduce selection bias and limit the generalizability of our results. Specifically, the selection of facilities to the clinical + payment group was based on several important characteristics, including a minimum CMS Star Rating, and were located in largely urban and suburban Central Indiana, may have resulted in bias and as compared to the payment only group. We lack a true control group and instead used individuals’ risk over time to account for within-person confounding as well as differences in the intervention facility to contrast risk. The lack of control for some potentially important variables, such as the presence of newly diagnosed terminal diseases, family/resident preferences, or other relevant unmeasurable variables, could result in omitted variable bias. In such a case, the attribution of unobserved factors such as a do not resuscitate orders could bias the associations of other characteristics. We note this as an alternative explanation for the lower risk of transfer among residents with advanced cognitive impairment. However, prior work within this same demonstration project found that associations between advance care planning and potentially avoidable hospitalizations were attenuated after accounting for facility clustering and resident characteristics.35