Pilot Testing the Johns Hopkins Early Discharge Planning Calculator

ABSTRACT

Objectives: To validate the Johns Hopkins Early Discharge Planning Calculator (JH-EDPC) machine learning algorithm in routine clinical practice for predicting postacute care (PAC) needs and evaluate whether knowledge of predictions influenced physical therapy (PT) consultation rates.

Study Design: Pilot controlled clinical validation study conducted from October 2022 to March 2023.

Methods: The JH-EDPC mobile app was deployed to 7 hospitalists (intervention group patients, n = 51 adult inpatients hospitalized ≥ 48 hours); hospitalists on the same units served as controls (patients, n = 111). Predictors included age, admission date, living status, surgery, and daily Activity Measure for Post Acute Care (AM-PAC) mobility scores. Accuracy metrics (area under curve [AUC], sensitivity, specificity, negative predictive value [NPV]) were calculated at a PAC probability cutoff of 0.25 using lowest 48-hour AM-PAC scores. Day-7 PT consultation rates were compared using inverse probability of treatment weighting.

Results: Overall, 11.7% of patients required PAC. The JH-EDPC correctly predicted discharge location for 85% of patients (AUC = 0.77), demonstrating 68% sensitivity, 87% specificity, and a high NPV of 95.4%. Adjusted day-7 PT consultation rates were 38% for intervention vs 35% for controls. Among patients predicted for home discharge, intervention patients received fewer PT consultations (19% vs 23%).

Conclusions: The JH-EDPC successfully integrated into clinical workflows with high specificity and NPV to rule out PAC needs. Findings suggest potential for optimizing hospital resource allocation by directing rehabilitation consultations to patients with the highest need.

Am J Manag Care. 2026;32(9):In Press

doi:10.37765/ajmc.2026.90008

Multidisciplinary early discharge planning reduces hospital delays.¹ Operational factors, including rehabilitation consultation timing, affect up to 20% of discharges, increasing length of stay and readmissions.² We previously published on a machine learning algorithm, the Johns Hopkins Early Discharge Planning Calculator (JH-EDPC), to predict postacute care (PAC) needs using electronic medical record (EMR) data.³ In this pilot study, we aimed to validate this algorithm in routine clinical practice to assess the accuracy of the JH-EDPC in predicting PAC. Secondarily, we checked for a signal of whether knowledge of predicted PAC influenced provider rehabilitation consultation.

METHODS

The JH-EDPC mobile app was deployed to 7 hospitalists (intervention) between October 2022 and March 2023; hospitalists on the same units served as controls. Eligible patients were adults hospitalized for 48 hours or more and alive throughout hospitalization. Providers used the JH-EDPC to calculate PAC probability by entering admission date, age, surgery, and daily Activity Measure for Post Acute Care (AM-PAC) Basic Mobility 6-Clicks Short Form scores.

Analysis Outcomes

Outcomes were patients’ discharge location and physical therapy (PT) consultation.

Statistical Analysis

Predictive accuracy was computed using the lowest AM-PAC score within the first 48 hours; home was defined as PAC probability of less than 0.25. Groups were compared using standardized mean differences. Inverse probability of treatment-weighted cumulative incidence curves for PT consultation was computed and compared using bootstrap CIs⁴ in RStudio (version 2023.06.1+524).

RESULTS

Overall, 19 of 162 patients (11.7%) went to PAC. Seven hospitalists used the JH-EDPC for 51 patients. The control group included 111 patients. The intervention group had more Black patients (61% vs 51%) and fewer men (41% vs 51%) (Table) than the control group. Otherwise, they were similar.

Predictive Accuracy

The JH-EDPC correctly predicted discharge location for 85% of patients, with an area under the curve of 0.77. Sensitivity (ie, correctly predicting PAC), and specificity (ie, correctly predicting home) were 68% and 87%, respectively. The model achieved a high negative predictive value (NPV) of 95.4%.

PT Consultations

By hospital day 7, the adjusted cumulative incidence of PT consultation was 38% in the intervention group and 35% in the control group (Figure, Panel A). Among patients predicted for home discharge, the adjusted consultation rate was lower in the intervention group (19% vs 23%; Figure, Panel B). Although no statistically significant differences (P = .29) were observed, this suggests potential for more consultations in the intervention group overall and fewer for those predicted to go home.

DISCUSSION

We successfully integrated a real-world discharge decision tool into clinician workflows. The JH-EDPC’s predictions were most helpful in ruling out the need for PAC with a high NPV.³ Although the study had a small sample, took place at a single site, included a short study window, and was not powered to detect a difference in rehabilitation consultation, these findings suggest a potential impact among patients projected for home discharge, warranting further investigation. Additionally, although inverse probability of treatment weighting (IPTW) balanced measured patient characteristics, the pilot’s reliance on a small group of self-selected hospitalists introduces potential clinician-level selection bias that IPTW cannot address.

Potential future implications of such a tool include enhanced discharge planning through early and accurate identification of PAC needs. Care managers, social workers, and rehabilitation therapists could use this tool to prioritize patients, initiate earlier planning, and expedite insurance approvals.⁵ Such proactive planning may mitigate delays and reduce hospital-associated disability from immobility.⁶

The algorithm is straightforward and could be built into EMRs to help provider decisions. Embedding the JH-EDPC directly would streamline its use and enhance its impacts. Full integration would allow for real-time data entry and access, improve usability, and potentially reduce discharge delays. Tools like the JH-EDPC represent promising strategies for optimizing hospital resources and improving patient outcomes through earlier, more accurate discharge planning.

Author Affiliations: Department of Physical Therapy, University of Nevada (DLY), Las Vegas, NV; Department of Physical Medicine and Rehabilitation, Johns Hopkins University (DLY, EHH), Baltimore, MD; Division of Hospital Medicine, Department of Medicine, Johns Hopkins University (RE, EHH), Baltimore, MD; Department of Biostatistics, Bloomberg School of Public Health, Johns Hopkins University (EC), Baltimore, MD.

Source of Funding: The authors have no funding to disclose.

Author Disclosures: The authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

Authorship Information: Concept and design (DLY, RE, EC, EHH); acquisition of data (DLY, RE, EHH); analysis and interpretation of data (DLY, EC, EHH); drafting of the manuscript (DLY, RE, EC, EHH); critical revision of the manuscript for important intellectual content (DLY, EC, EHH); statistical analysis (DLY, EC); administrative, technical, or logistic support (DLY); and supervision (DLY, EHH).

Address Correspondence to: Daniel L. Young, PT, DPT, PhD, University of Nevada, 4505 S Maryland Pkwy, Box 453029, Las Vegas, NV 89154-3029. Email: daniel.young@unlv.edu.

REFERENCES

1. Wong EL, Yam CH, Cheung AW, et al. Barriers to effective discharge planning: a qualitative study investigating the perspectives of frontline healthcare professionals. BMC Health Serv Res. 2011;11(1):242. doi:10.1186/1472-6963-11-242

2. Micallef A, Buttigieg SC, Tomaselli G, Garg L. Defining delayed discharges of inpatients and their impact in acute hospital care: a scoping review. Int J Health Policy Manag. 2022;11(2):103-111. doi:10.34172/ijhpm.2020.94

3. Young DL, Hannum SM, Engels R, Colantuoni E, Friedman LA, Hoyer EH. Dynamic prediction of post-acute care needs for hospitalized medicine patients. J Am Med Dir Assoc. 2024;25(7):104939. doi:10.1016/j.jamda.2024.01.008

4. Xu S, Ross C, Raebel MA, Shetterly S, Blanchette C, Smith D. Use of stabilized inverse propensity scores as weights to directly estimate relative risk and its confidence intervals. Value Health. 2010;13(2):273-277. doi:10.1111/j.1524-4733.2009.00671.x

5. Capo-Lugo CE, McLaughlin KH, Ye B, et al. Using nursing assessments of mobility and activity to prioritize patients most likely to need rehabilitation services. Arch Phys Med Rehabil. 2023;104(9):1402-1408. doi:10.1016/j.apmr.2023.03.018

6. Zisberg A, Shadmi E, Gur-Yaish N, Tonkikh O, Sinoff G. Hospital-associated functional decline: the role of hospitalization processes beyond individual risk factors. J Am Geriatr Soc. 2015;63(1):55-62. doi:10.1111/jgs.13193