"Shopping Around": Preferences and Incentives for Long-term Placement
Contents
Introduction
In the US healthcare system, a small fraction of patients account for a substantial portion of post-acute hospital stays: although 2% of patients spend more than 21 days in hospital, this proportion accounts for approximately 15% of all patient days in hospitals [invalid citation] [invalid citation] [invalid citation] [invalid citation]. Over half these days are post-acute, where a patient has been treated for their acute care needs and transitional planning is completed [invalid citation] [invalid citation].
Research suggests that a leading cause of post-acute discharge delays is finding available beds in community-based care homes that offer nursing-home level of care [invalid citation], with decision making among stakeholders, including care home operators, contributing significantly in time spent discharge planning [invalid citation] [invalid citation].
In this paper, we discuss how patient placement can be viewed as a digital marketplace [invalid citation] [invalid citation]: hospital staff use a computer-mediated system to find care homes that match patients in need of placement, and care home operators use the same system to find patients to fill vacancies. Based on this framing, we describe an experiment to collect and refine care home operators' preferences, with the goal of facilitating better matching and improving patient placement.
Research Context
Our study is based on a three-year collaboration with a major hospital in Hawai‘i, deploying an SMS-based system to facilitate long-term patient placement. This system aids care homes in filling vacancies and assists hospital staff in finding suitable accommodations for post-acute patients.
Motivational Analysis
We first analyze historical data from our system to understand patient placement challenges. We begin with a regression analysis on data from 760 patients, correlating the length of hospital stay with their mobility levels (assessed by a nationally recognized ambulation score). Results indicate that homes often shy away from less mobile patients despite higher insurance payouts, suggesting that current incentives are insufficient. We subsequently contextualize our regression analysis findings with how care homes express preferences on our platform.
We explain how a quarter of homes explicitly state preferences, concerning both patient conditions and general characteristics like sex and weight. Other homes express no preference, ostensibly accepting any patient; however, in practice, they often do not accept any patient but in fact, are evaluating multiple options, a phenomenon described as "shopping around." We categorize these behaviors as "underconstrained," where the expressed lack of preference does not reflect genuine interest, undermining efficient placement. Conversely, some homes are "overconstrained," specifying rare ideal patient profiles, leading to prolonged vacancies, when they would accept other patients. Our research aims to refine preference assessment methods to improve data quality and system efficacy in patient placement.
Randomized Controlled Trial
Phase 1
We conducted a two-phase randomized controlled trial (RCT) to determine whether expressed preferences reflect ground truth or are malleable. This study involved 960 care homes, equally split into control and treatment groups, with the treatment group further divided based on homes' preferences being potentially overconstrained or underconstrained. Phase 1 of our experiment evaluated preference malleability. Homes in the control group received routine messages confirming their current preference status and prompting updates if necessary. By contrast, the overconstrained homes in the treatment group received tailored messages encouraging them to reconsider stringent preferences, suggesting that they might receive more matches if they do; underconstrained homes were asked whether they were interested in a patient that matched their preferences but who matched the least with other homes, which we describe as "hardest to match" patients.
Phase 1 indicated preference malleability; homes in the treatment group were over 50% more likely to change their preferences (p-value = 0.04), suggesting some flexibility when confronted with alternative scenarios. Homes were more willing to adjust their weight preferences than sex preferences; the rate of sex preference changes was not statistically significantly different between control and treatment, while weight showed borderline significance (p-value = 0.054).
Phase 2
Phase 2 presented all homes with profiles of their hardest to match patients, to gauge interest and whether Phase 1 treatment messages increased matches. Phase 2 did not reveal significant differences in patient interest between the control and treatment groups: both groups demonstrated similar levels of engagement when presented with actual patient profiles. However, engagement among homes was notably higher when homes were provided with specific patient scenarios that closely matched, or otherwise challenged their stated preferences.
We thus emphasize the importance of precise and contextual communication in eliciting meaningful responses from care homes. Overall, our RCT highlights the complexity of preference management in patient placement systems and suggests that while homes may be open to modifying their preferences, significant shifts are contingent on clear, compelling communication that directly relates to their operational realities and patient care capacities.
Interviews
To deepen our understanding of our experiment results, we conducted interviews with 22 participating care homes from both the control and treatment groups. We show how homes' preferences are shaped by a combination of economic need, load balancing, and operational risk. For example, we discuss how a home may have a given roster of patients around whom they must align new admits (e.g., a patient may not want to share a room with a patient of another sex). We also see how long-standing biases around patient aggression further contribute to differing malleability between sex and weight preferences.
Discussion
We conclude by discussing how fundamental marketplace characteristics are facilitated through the preference solicitation and refinement mechanisms. We also highlight challenges and implications for systems that aim to collect and manage fine-grained, changing preferences to facilitate matchmaking in high-stakes contexts like patient placement.
Summary Table
| Stage | Research Question | Finding |
|---|---|---|
| Motivational Analysis | How does ambulation impact duration to placement? | The more complex the patients' needs, the harder to place. |
| Motivational Analysis | What preferences do care homes indicate and how do they impact patient matches? | Some homes have preferences, but many others do not state their preferences. Most homes' preferences shared on the system involve sex or weight, though some are more complex. |
| Would homes benefit from suggestions to change preferences? | Homes may be over- or under-constrained, and would benefit from changing preferences. | |
| Randomized Controlled Trial | To what extent are homes’ expressed preferences around patient sex and weight malleable? | Homes change their preferences when probed to do so, and are more open to changing weight preferences than sex. |
| Do refined preferences on our system improve homes’ interest in patients that match their preferences? | Probing did not show a significant effect on patient interest, though it increased engagement. | |
| Interviews | What factors might explain the differences we observed in how weight and sex preferences change? | Economic needs significantly influence preferences related to ambulation and weight, but not sex. Care homes must balance staff capabilities, patient needs, and roster composition, which influences weight-related preferences more than sex-related ones. Social desirability affects preferences, particularly for patient aggression, which strongly influences sex preferences over long time horizons. |
| What factors might explain the absence of observable differences in patient interest between treatment and control groups? | SMS design constraints, including expectations about texting and the inability to capture conditional preferences, may explain why patient interest remains unchanged over SMS. |