Assessment of Patient Ambulation Profiles to Predict Hospital Readmission, Discharge Location, and Length of Stay in a Cardiac Surgery Progressive Care Unit

Key Points Question Are patient ambulation profiles predictive of hospital readmission, discharge location, and length of stay? Findings In this prognostic cohort study of 100 adults in a cardiac surgery progressive care unit, patient ambulation profiles were predictive of 30-day readmission (C statistic, 0.925), discharge location (C statistic, 0.930), and length of stay (correlation coefficient, 0.927). Meaning Patient ambulation profiles from a real-time location system enable prediction of clinically relevant outcomes.


eAppendix 1. Study Criteria and Patient Flow Chart.
Study criteria: patients who underwent surgery from 8/29/16 to 4/4/18 Inclusion criteria included individuals undergoing coronary artery bypass graft surgery (CABG) and/or valve surgery, aortic surgery, and/or myectomy. Exclusion criteria included lung or heart transplant, or planned insertion of ventricular assist device. To account for the fact that the transfer day and discharge day usually reduced the opportunity for ambulation, parameters were calculated based on the total number of days on the PCU and the total number of days excluding the transfer and discharge days (full days). Therefore, each ambulation profile contains 19 ambulation parameters + 1 length of stay parameter = 20 parameters for either the total number of days or full days on the PCU. These parameters are defined below.

Patient Flow Chart
1. Total number of ambulations, N. The total number of ambulations completed during a patient's stay on the PCU. An ambulation is arbitrarily defined when a patient's badge is detected by at least 6 ceiling sensors in the corridor before returning to the patient's room. As for all parameters below, the total number of ambulations is calculated for total number of days or total number of days excluding transfer and discharge days.  Derivatives 17. Change in ambulation distance (dn/n). The slope of ambulation distance versus ambulation number during the patient's stay on the PCU. n = 1, 2, …, N. The slope was determined from a linear least squares fit. 18. Change in ambulation duration (tn/n). The slope of ambulation duration versus ambulation number during the patient's stay on the PCU. The slope was determined from a linear least squares fit. 19. Change in ambulation mean speed (sn/n). The slope of ambulation speed versus ambulation number during the patient's stay on the PCU. The slope was determined from a linear least squares fit.

Ambulation frequency (N/LoS
Other 20. Length of stay, LoS. The total number of days on the PCU. The number of full days on the PCU is LoS -2 (transfer and discharge days).
Abbreviation key nambulation number (n = 1, 2, …, N) Ntotal number of ambulations dambulation distance (ft) tambulation time (minutes) sambulation speed (m/s) days and completed 2 voluntary ambulations on day 1, three ambulations on day 2, three ambulations on day 3, and 1 ambulation on day 4 prior to discharge. Note that this individual was on the PCU for two full days. The PCU floor plans show the path of each ambulation. Each ambulation is comprised of individual segments determined from the time stamp and location that a patient badge is detected by a ceiling sensor. The cumulative distance for each ambulation increases monotonically with time indicating that the patient walked continuously at an approximately constant speed without resting. The total distance for each ambulation increased from about 400 ft on day 1 to almost 800 ft on day 4. Note that the average speed per ambulation increased progressively from about 1 mph on day 1 to about 1.5 mph on day 4. See eTable 1 for the ambulation profile for this patient.
number of days and full days (total days excluding transfer and discharge days) on the PCU for patient 83. This individual was discharged to home and was not readmitted. The derivatives represent a linear least squares fit to a plot of ambulation distance, ambulation duration, and mean ambulation speed versus time during an individual's stay on the PCU. number of days and full days (total days excluding transfer and discharge days) on the PCU for patient 92. This individual was on the PCU for 6 days and was readmitted 13 and 27 days after discharge. Note that the patient did not complete any ambulations on the transfer and discharge days, and hence all ambulation parameters except ambulation frequency and compliance are the same for total days and full days. The derivatives represent a linear least squares fit to a plot of ambulation distance, ambulation duration, and mean ambulation speed versus time during an individual's stay on the PCU.

eTable 3. Summary of Ambulation Parameters for 100 Patients (Total Days on the PCU).
Analysis based on ambulation data for the total number of days on the unit. Data represent mean (± SD).

30-day Readmission
Discharge to Acute Rehab.

eFigure 3. Comparison of Ambulation Parameters for 30-Day Readmission and Discharge to Acute/Subacute Rehab.
Bars represent mean ± 95% confidence limits. Comparison based on total days on the unit. 0 -not readmitted or not discharged to acute/subacute rehab. 1 -readmitted or discharged to acute/subacute rehab. N = 100.

eTable 5. Summary of F and P Values for Comparison of Outcomes Between Groups (Readmission Yes/No; Discharge Location Home/Acute Rehab) for Each of the 19 Ambulation Parameters and Length of Stay. Ambulation data for 100 patients (total days on the PCU).
Threshold values were determined in the following way. If there was no overlap of the ±95% confidence limits between the two groups then the threshold was determined from the average of the lower and higher values of each group. If the bands overlapped then the threshold was determined as the average of the mean values of the two groups.
F and p values for ambulation parameters associated with 30-day readmission (yes/no) and discharge location (yes/no):

eFigure 4. Examples of Ambulation Distance (dn), Time (tn), and Speed (sn) Versus Ambulation Number (n).
The derivatives dn/n, tn/n, and sn/n, provide information on the changes in a patient's mobility during recovery on the PCU. The slopes of these plots, determined from a linear least squares fit ,were included in each patient's ambulation profile. (A) Profiles for patients P7, P74, and P83 (not readmitted within 30days), (B) profiles for patients P37, P70, and P92 (readmitted within 30 days of discharge). P7 completed twelve voluntary ambulations but showed no obvious trend in terms of distance. However, the duration of each ambulation decreased corresponding to a monotonic increase in ambulation speed. P74 completed six ambulations but with only a small increase in distance over time. However, the decrease in duration also resulted in a monotonic increase in average speed. P83 completed nine ambulations with a monotonic increase in distance in approximately the same duration, corresponding to an increase in average speed. The three individuals who were readmitted within 30 days showed small increases in distance and duration, but the corresponding average speed was either constant or decreased with number of ambulations.

eFigure 5. Comparison of Ambulation Distance (dn), Time (tn), and Speed (sn) Versus Ambulation Number (n) and Time on the PCU (Beginning at 00:00 on the Transfer Day).
The changes in dn, tn, and sn can be defined in terms of the number of ambulations (dn/n, tn/n, and sn/n) or in terms of the time on the unit (dn/t, tn/t, and sn/t) beginning at 00:00 on the transfer day. To assess whether there was a difference, we determined the derivatives for both the ambulation number and time on the PCU. (A) Distance, duration, and average speed versus ambulation number and time on the PCU for patient 83 who was discharged to home and was not readmitted within 30 days. (B) Comparison of change in ambulation speed determined from ambulation number (sn/n) and time on the PCU (sn/t) for all patients (N = 100) based on whether they were readmitted within 30 days.
Bars represent mean ± 95% confidence limits. The correlation between non-readmitted and readmitted patients with change in speed per ambulation (sn/n) is significantly larger than for change in speed per time on the PCU (sn/t). Therefore, change in speed per ambulation (sn/n) was used in the ambulation profiles and predictive models.

eTable 6. Comparison of Prediction Models for 30-Day Readmission and Discharge Location Based on Total Days and Full Days on the PCU. Models based on 19 parameters in ambulation profile.
Having established that several ambulation parameters were statistically significant between 30-day readmission and discharge location groups, a stepwise binary regression model was used to assess predictions using a split sampling method (70/30). In the model based on total days, 30-day readmissions were predicted with 86.7% sensitivity and 88.2% selectivity and the discharge location was predicted with 84.6% sensitivity and 86.4% selectivity. The model predictions for 30-day readmissions improved slightly (93.3% sensitivity and 88.2% selectivity) based on full days on the PCU, but decreased for discharge location (84.6% sensitivity and 77.3% selectivity). Full days represents the number of days on the PCU excluding transfer and discharge days. Since the number of 30-day readmissions and cases of discharge to acute rehab were relatively small, we used a synthetic minority over-sampling technique (imblearn.over_sampling.SMOTE, Python) to balance the data prior to analysis. PREDICTORS 30-day readmissions: LoS on PCU, ambulation frequency, number of days with ambulations, % days with 1 ambulation per day, % days with 3 ambulations per day, total distance, mean ambulation distance (all ambulations), shortest ambulation distance, maximum mean speed in a single ambulation, mean speed (all ambulations), change in ambulation duration (tn/n). Discharge to acute rehab: ambulation frequency, number of days with ambulations, % days with 1 ambulation per day, % days with 2 ambulations per day, total distance, longest ambulation distance, mean ambulation distance (all ambulations), shortest ambulation duration, mean ambulation duration, mean speed (all ambulations), change in ambulation distance (dn/n) Note. Predictions based on the binary stepwise regression method, including or removing an independent variable at each step based on the p-value of F (Entry: 0.05, Removal: 0.10).