Children’s Hospital of Pittsburgh Tertiary care facility 315 beds 41 Bed ED 103 critical care beds 13 OR suites 40+ specialties
Main campus – Lawrenceville Four Ambulatory Care Centers Seven Children’s Express Care locations Five Specialty Care Centers
Children’s Hospitals are generally considered as not competing with ‘adult’ hospitals in their region
Patients experience a long wait time to make appointments. The services state that they cannot increase capacity because of lack of space provided by CHP for appointments. Space is currently allocated based on doctor monthly schedules. CHP observes that rooms are not fully utilized during the day. Can CHP predict room availability or additional needs to reallocate rooms between services?
For each session, compare rented grid and the staffed rooms as determined from the patient time in room. Some sessions require more room than rent grid, but generally rent grid is greater than staffed rooms.
Demographics of patient Gender, age, zip code, insurance status, race/ethnicity New/return Schedule Scheduled provider time Actual appointment Status: Completed, Cancelled, No-show Patient in room time, Discharge time
“Rented” rooms – For each session (AM/PM), rooms allocated to service for use in clinic Actual room use Not currently recorded Derive this from the patient time in room Determine the maximum number of patients in rooms simultaneously during a session Verified method through one month of observation
Actual room utilization based on patient time in room is low. Suggests that rooms can be more effectively assigned. Rented rooms do not account for variations in doctor schedules.
Goal: develop a predictive model that predicts actual room use given the current state of the schedule. 2 days ahead, 1 week ahead, 2 weeks ahead CHP can identify services that are not expected to use all of their rented rooms, and identify rooms for reallocation to services that have additional patients.
Not all patients have rooming or discharge times recorded. Impute missing data by estimating patient time in room conditioned on type of patient (new/return) and the scheduled appointment time. If both rooming time and depart summary are unavailable, impute the rooming time based on the scheduled appointment time, patient type and scheduled provider time.
Scheduled room use - Estimate patient time in room based on patient time, scheduled appointment length, and schedule appointment time. Find maximum scheduled patients in room simultaneously. Demographics of scheduled patients Schedule status at time horizon EHR data includes when appointments are made or cancelled.
For each model family, use 10 fold cross validation and show RMSE for 1 week look ahead. RMSE vs complexity parameter graph for the one week model
Table with six models and RMSE for 1 Week look ahead.
Models will predict staffed rooms at two weeks, one week, and two working days ahead.
Develop another set of models for 2 day ahead prediction Repeat analysis for all services Determine how to handle larger proportion of missing data cases Develop policy for reallocating rooms based on the predictive model Allow for a safety room