Introduction

Many countries are currently experimenting with bundled payment (BP) models, which prospectively finance a specific set of services included within a standardized care pathway [1]. From the perspective of the health authorities, the main goal of a BP model is to promote the coordination of healthcare professionals in the care pathway, which is considered a possible means of improving quality of care.

The French Social Security Financing Act for 2018 included a national initiative to test organizational innovations through different financing models. Three major experimental models are still being tested for hospital or ambulatory care. Among those, “Episode of Care” aims to introduce a flat BP rate for voluntary hospital and community-based ambulatory care (CBAC) professionals involved in the same standardized surgical care pathway (both before and after the surgery). For each episode of care, the lump sum takes 3 components into account: 1) a basic package covering the initial stay (hospitalization for surgery), 2) an adjustment by defined modulators (the presence of digestive or cognitive comorbidities, or chemotherapy treatment preceding the surgical act for example), and 3) an amount covering the average risk of hospital readmission after discharge during a defined period. Therefore, no additional funding is provided if the patient is readmitted for complications following their initial stay. BP models could thus contribute to reducing readmission rates by providing a target for hospitals to achieve [2]. In addition, health professionals are financially incentivized to participate through the potential distribution of funds gained from avoiding costly readmissions.

To help hospitals meet their readmission rate reduction goals, it is essential to consider the factors that influence readmission. It appears that readmissions are closely linked to the length of stay (LOS) of the index stay [3], which is why there is extensive literature on the link between optimal LOS and readmission. Based on different statistical models [4,5], this relationship is often described as a J-shaped curve: for a very short LOS, as the LOS increases, the readmission rate decreases, up to a threshold. Then, after this threshold, the readmission rate increases as the LOS increases. Many factors have been shown to influence the role of LOS on readmission, such as patient health status [6,7]. For example, for most conditions, increasing the LOS increases the readmission rate for healthy patients, whereas increasing the LOS decreases the readmission rate for less healthy patients. However, this is not always the case, such as in orthopedic surgery (total joint replacement), for which readmitted patients have a longer index LOS regardless of comorbidities [8]. Thus, optimizing LOS to reduce the readmission rate, and thus avoiding treating patients a second time within the same BP, should take into account both the patient’s condition and the type of surgical procedure [9].

An large-scale evaluation of experiments implemented in the USA showed that the length of the initial stay could be a lever for lowering readmissions and ultimately reducing the overall cost of the treatment [10]. Hospitals could therefore consider that the decrease in the readmission rate offsets the marginal cost of an additional day of hospitalization during the index stay, which is an incentive to increase LOS.

The new care pathways introduced in France starting in 2015–2016, notably the Enhanced Recovery After Surgery (ERAS) care programs, have shown that improving the organization of care can also reduce LOS [11]. ERAS hospitals were identified and certified by the GRACE network. This certification ensured that the recommendations of the French health authority were followed. ERAS is a comprehensive approach to patient care that promotes early recovery from surgery. It is part of a hospital project and is based on a clinical pathway that includes the before, during and after phases of surgery. Implementing such a program represents a step towards improving practices across all teams. It requires a reorganization of care and concerted efforts within a multi-professional team involving all of the healthcare professionals involved in inpatient and outpatient care. ERAS is based on informing patients about their care pathway, anticipating and organizing their care and discharge, avoiding the consequences of surgical stress, managing pain in all situations, and stimulating patient autonomy. ERAS programs have been shown to decrease LOS without increasing readmissions [12,13]. A feasibility study published in January 2014 was conducted in five pilot sites, demonstrating the interest and feasibility of ERAS in the French context [14]. The promising results obtained abroad with these new practices encouraged French healthcare facilities to experiment at the time. Outside the context of French BP experiments, other programs using the LOS and the readmission rate as indicators of quality of care have already been evaluated. For example, multidisciplinary discharge rounds have already been shown to reduce LOS and readmissions [15].

In this study, we wanted to focus on patients with a given condition undergoing a specific surgical procedure to analyze the relationship between the index LOS and readmission rates. We know that readmission, which is controllable to various degrees, is important for practitioners because it is a partial reflection of the efficiency of care, all other things being equal. The objective of this illustrative situation was to determine whether novel care pathways which seek to optimize care or coordination between healthcare professionals during the initial stay can reduce readmission rates through changes in LOS in the context of BP. We chose the case of cancer patients undergoing colectomy, which is a pathway that has been defined in French BP experiments and for which the readmission rate was highest in the experiments. Given the possibilities of immediate application in the field, we believe that evaluating this LOS/readmission relationship is particularly relevant. In order to make our interpretations more relevant, we based our investigations on the data used by the Ministry of Health to construct the BP pathway. Our results therefore make it easy to identify potential changes that could be made to the pathway to improve the readmission rate.

Methods

Results

Description of the population and hospital stays

The overall population characteristics are presented in Table 1. Patients 60 years or younger were the smallest portion of the study population (16%). Most patients had no chemotherapy following surgical treatment (78%), no other comorbidities (90%), and no cognitive disorders (97%). Males were more likely to be readmitted than females (53% vs. 47%).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Patients treated for cancer by colectomy in France, 2014-2016.

https://doi.org/10.1371/journal.pone.0346558.t001

The characteristics of the hospital stay (42,603) are described in S1 Table. The DRG reflects disease severity (level 1 being the lowest level of severity and 4 the highest). This variable was constructed by the Ministry of Health and is used to modulate hospital funding. Patients were most frequently classified as severity level 3, but almost 55% of the total were severity levels 1 and 2. Most patients were hospitalized in private (46.6%) or public (42.4%) hospitals. Only 9.1% of patients were readmitted after their index stay.

The mean LOS was 12.5 days, and the median LOS was 10 days, with extreme values ranging from 0 to 321 days in the overall population. The mean LOS was lower in the ERAS subgroup (11.4 days). When analyzed by level of severity, the median LOS was almost identical in the public and private sectors (S2 Table).

Analyses depending on the implementation of ERAS showed that readmissions were lower in ERAS subgroup (7.9% for ERAS vs 9.3% for others, p < 0.01), as were cognitive disorders (2.3% for ERAS vs 3.2% for others, p < 0.01). However, digestive disorders were more frequent (15% in ERAS subgroup, vs. 12% for others, p < 0.001), and the proportion of patients covered by the CMU was lower (0.6% in ERAS subgroup, vs. 1% for others, p = 0.03) (S3 Table). There was no significant differences between groups in terms of socio-environmental difficulties (p > 0.9), age group (p > 0.9), gender (p = 0.5) or chemotherapy treatment (p = 0.08).

Distribution of readmission

Readmission occurred in around 9% of all healthcare pathways (S1 Table). Using a segmented regression to study the relationship between LOS and readmission, we obtained a J-shaped curve (Fig 1). First, we observed a significant decrease in the readmission rate for LOS between 0 and 5 days. Then, after the breakpoint (around 5 days), we saw a progressive increase in readmissions as the initial LOS increased.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Readmission rate and Length of Stay (LOS), for colectomy pathway in France, between the 1^st January 2014 and 30^th June 2016.

Note: the blue curve corresponds to the value of the readmission rate calculated with its confidence interval by glm modelling; the black curve corresponds to the value calculated by segmented regression with the estimated breakpoint.

https://doi.org/10.1371/journal.pone.0346558.g001

The shape obtained was nearly the same whatever the level of severity defined by the DRG and the sector (public or private). However, the curves appear flatter for levels 3 and 4 (S2 Fig). The breakpoint was around 7 days for the level 1 group.

Table 2 shows the segmented multivariate regression results. Taking into account available confounding factors, we showed a significant effect of LOS, whose coefficient was negative. Overall, the estimate breakpoint was around 5 days. We observed that, before the breakpoint, increasing the LOS variable decreased the probability of readmission (coefficient = −0.016, 95% CI [−0.011; −0.021], p < 0.01). The main results were confirmed by sensitivity analyses (S4–S6 Tables). The results were similar for the chemotherapy and gender variables. However, the digestive disorders variable increased the probability of readmission (coef. = 0.022, 95% CI [0.018; 0.028], p < 0.001).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Segmented regression modeling the probability of readmission as a function of the control variables used to calculate the bundled payment experiment in France.

https://doi.org/10.1371/journal.pone.0346558.t002

For the LOS subgroup of less than 7 days, the estimated breakpoint was 4 days and coefficient was also significantly negative (coef. = −0.034, 95% CI [−0.043; −0.026], p < 0.001) (S7 Table). For the ERAS subgroup, the estimated breakpoint was also around 4 days and the LOS variable remained weakly significant (coef. = −0.026, 95% CI [−0.040; −0.012], p < 0.1) (S8 Table).

Discussion

This work raises questions about the potential relationship between LOS and readmission, and about future practices in the context of French BP experiments. Taking into account the use of the ERAS method, which is on the way to becoming a best practice, using among other things minimally invasive surgery, we found that LOS and readmissions were lower in the ERAS subgroup. These new care pathways significantly reduced LOS and the value of the breakpoint. Previously published analyses by the RAND Corporation [1] showed that bundled payments could align financial incentives to reduce LOS and readmissions, fostering greater coordination of care. Consistent with our hypothesis, a study by Westert et al. [10] noted similar effects in countries such as the U.S. and the U.K., where readmissions were integrated into capitated or bundled frameworks. These findings support the idea that BP initiatives can improve both efficiency and quality, although robust oversight is needed to prevent cost-shifting or selective patient enrollment.

While it is not directly confirmed in our work, we may wonder whether ERAS could have a positive effect on readmission rates. The ability of such initiatives to improve coordination and multidisciplinary management is not always conclusive, and questions remain about their impact on readmission rates at 30 days, 90 days, or even 6 months [21].

Given the contrasting results obtained thus far, and considering that each experiment and the coordination methods are different, it is likely that the results of the French initiatives will also have contrasting results for readmissions. In addition to improving coordination, the availability of healthcare professionals, particularly nurses, is essential, and we now know that this factor must also be taken into account [8]. Before implementing an experiment, each hospital must assess its practices, its case mix and its means of coordination and multidisciplinary management to better estimate the potential decrease in readmissions and to maximize their gains with the BP initiative. Indeed, modifying practices is a means of optimizing costs and patient outcomes in health establishments [22]. In view of our results, BP experiments could promote new practices such as ERAS, which will maximize the gain for hospitals and perhaps even improve the quality of care, although this point remains to be proven.

Thus, while LOS was shorter in the ERAS group during our study period, it remains very long compared with current LOS, which is closer to 5 or 6 days [23,24]. There are probably several reasons for this result. Firstly, we used data from the start-up period of ERAS in France in 2015. Thus, the LOS observed in this program were similar to those seen in the usual care pathways. In addition, our work was based on real-life data and not solely on data from expert centers. Finally, not all ERAS centers necessarily included patients in their protocol. The ERAS patient group had a better socio-economic status and fewer cognitive disorders in the pre-operative period, which is likely because ERAS involves controlling for the known risks of readmissions before surgery. On the contrary, digestive disorders were more frequent in the ERAS group, suggesting that patients were in poorer health before surgery and presented a higher risk of complications afterwards. This could suggest that ERAS does not encourage the selection of the least severe patients, which would have led to a selection bias.

On average for all groups, the observed LOS was higher than the estimated breakpoint. Best practice could therefore aim to reduce LOS in order to reduce readmissions, whatever the level of severity. The estimated breakpoint was between four and five days, depending on severity but also depending on practices, which suggests a possible target LOS. Moreover, our study showed that a longer initial stay up to the observed breakpoint (short stays), led to a lower readmission rate in the colectomy pathway due to the weight of LOS within the model, whose coefficient was always negative and significant. This conclusion did not apply to patients whose LOS was already high, either because they had several comorbidities on admission, or because they developed complications during their stay. We found that increasing LOS after the breakpoint was associated with an increased risk of readmission, and that patient characteristics such as digestive disorders or other comorbidities outweighed the effect of LOS. These results are consistent with previous studies that have described similar observations for other medical or surgical procedures [6,25–28]. Indeed, digestive comorbidities significantly increase both readmission rates and length of stay for patients undergoing partial colectomy [2,6]. In contrast, adjuvant chemotherapy appears to be associated with a meaningful reduction in postoperative readmissions [3], whereas higher hospital-level severity, measured by DRG classification, further exacerbates these outcomes [7]. One hypothesis is that, for those patients whose postoperative course was uncomplicated, adjuvant chemotherapy was helpful.

On the whole, we underline the impact of these kind of organizational innovations. In the United States, BP programs have been linked to a reduction in average surgical length of stay [29], while DRG-based incentives in the UK similarly corresponded to a long-term downward trend in LOS [30]. These findings align with evidence that ERAS protocols independently reduce LOS by approximately 1–2 days [31,32], further supporting the synergy between financial reform and enhanced perioperative practices.

Conclusions

This work is part of the ex-ante evaluation of innovative financing methods within the framework of the BP initiative in France, particularly following the launch of the BP project. Consistent with shorter stays obtained with ERAS (4–6 days), our findings suggest that BP could similarly encourage hospitals to optimize LOS around the 5-day breakpoint and thereby minimize readmissions. This is a preliminary finding that does not prove a causal relationship. Nevertheless, it would be worthwhile to explore this issue further in future research.

Before the post-implementation evaluation of this new initiative, there is a need to explain the current performance of the healthcare system in the specific case of the care pathways chosen for the BP. This data can then be used to clarify the current pattern of care and highlight possible efficiency gains. Our results confirm that LOS is a relevant indicator that could be used to monitor changes in practice while seeking to improve hospital readmission rates. This is particularly interesting in the context of the BP financing model, which encourages the use of novel care pathways such as ERAS.

In light of these findings, policymakers can consider three major recommendations to enhance care efficiency and outcomes. First, broader adoption of ERAS protocols can be encouraged through targeted funding and training programs, given their association with shorter LOS and fewer readmissions. Second, the hospital could consider LOS as a potential lever for reducing readmission rates, and thus implement clinical research to focus on this association.

Finally, it is crucial to adjust bundled payment models for patient complexity – including digestive comorbidities – so hospitals treating more severe cases are not penalized.

In this work, we did not study rehabilitation pathways after the initial hospital stay. Further research is needed regarding the impact of these pathways on readmission rates, clarifying the potential effects of BP.

Supporting information

Acknowledgments

We would like to thank Technical Agency for Information on Hospital Care (ATIH) for the availability of the data and for their assistance. The authors thank Suzanne Rankin (Dijon Bourgogne University Hospital) for providing English language editing. We would like to thank the GRACE association for its invaluable support.