LEADING INDICATORS TO CUT PATIENT STAYS.
One challenge with using length of stay as a key performance metric is that it's a lagging statistic; it is known only after a patient has been discharged. Other factors--including social determinants of health, availability of post-discharge resources, family support, transportation availability and specific institutional practices--can affect stay lengths as well.
From a daily-operations perspective, it's important to find good leading metrics that front-line staff can influence in real time to improve the patient experience and help reduce stay lengths. Several best practices suggest moving discharges to earlier in the day to decrease emergency department boarding hours and the overall throughput of patients. (4,5) However, to date, there is no evidence that such practices are associated with lower lengths of stay. In contrast, one study looked at discharges before noon and found longer stay lengths with higher performance in this metric, especially for patients admitted from the ED. (6)
The purpose of our study was to evaluate the correlation of inpatient lengths of stay and boarding hours with three potential leading metrics: provider discharge orders by 10a.m., discharge orders to out-of-hospital within three hours, and a combined discharge goal of patients out of the hospital by 1 p.m.
This is a retrospective review of 83,899 adult and pediatric inpatient discharges at an academic medical center from July 1, 2014, to Dec. 19, 2016. This population included psychiatric patients and all patients discharged from general medical/ surgical units, including obstetrics. Length of stay was defined from the moment an admission order was placed until the patient left the hospital. All patients with a stay length of less than 48 hours were excluded from the analysis, since a short stay length might conflict with the principle of discharges earlier in the day. Likewise, long lengths of stay (greater than 20 days) also were excluded to prevent outliers from skewing the distribution.
Starting in July 2014, clinical leaders were encouraged to focus on improving stay length.
Physicians were encouraged to get at least 40 percent of their discharge orders written by 10 a.m. The percentage was chosen to smooth out discharges throughout the day, as studies have demonstrated focusing most discharges during one period of time leads to delays in care. (7) The time fit well with the completion of morning rounds for most specialties.
Nurses had a goal of 70 percent of patients to be discharged from the hospital within three hours of the order being placed. This was based on baseline metrics with a reasonable stretch goal set beyond this, of within 5 percentage points each that were determined to be statistically probable to achieve.
After a year of evaluation, feedback from staff led to concerns of physician orders being placed before a patient was medically ready for discharge, leading to an unattainable three-hour window. Conversely, concerns were raised by providers that nurses discouraged providers to place the discharge order until later to help them reach the three-hour window.
To create better alignment between the two groups, a combined goal was created: 28 percent (40 percent multiplied by 70 percent) of all patients should leave the hospital by 1 p.m. (10 a.m. plus three hours).
The analysis for the study looked at the correlation of operational length of stay, defined as the time from admission to discharge, with the three leading metrics. In addition, a correlation with the length of stay index, defined as observed over expected LOS, was performed. Expected LOS is risk-adjusted based on discharge diagnoses and patient acuity. (8) Finally, we looked at the correlation of ED and post-anesthesia care unit boarding times in terms of minutes. Given the large number of psychiatric boarders in the ED, a secondary analysis was done for this patient population. For other clinical services with admitting privileges, their sample sizes were too small to conduct similar sub-analyses with sufficient statistical power.
Log-linear regression was used for the LOS index to ensure results weren't skewed by outliers. In the case of boarding minutes and observed LOS, Poisson regression was used to appropriately match the distributions. Cook's distance values greater than 1 were removed, and regressions were rerun to ensure high leverage points were removed from the models. Coefficients were reported as exponentiated so they can be interpreted as a multiplicative increase. P-values of less than 0.05 were considered statistically significant.
The age, sex and race distribution of the patient population are demonstrated in tables 1-3. The majority of patients are under the age of 12, female and Caucasian. The top 20 diagnosis-related groups of patients are listed in Table 4, with psychoses the most common diagnosis. The actual length of stay was longer for those admitted with a psychiatric diagnosis vs. those without, in Table 5.
For the following five paragraphs, all p-values are less than 0.05 unless noted.
When looking at the operational stay length, 3.79 percent of the variance could be explained by the three leading metrics. Discharges completed within three hours of the discharge order were associated with a 9.4-percent reduction in operational stay length. In contrast, having discharge orders written by 10 a.m. correlated with a 3.8-percent increase in operational stay length. Although patients discharged by 1 p.m. correlated with a 1.1-percent increase in operational length of stay, the relationship was not statistically significant (p-value equals 0.2).
For the length of stay index, 1.7 percent of the variance could be explained by these three leading metrics. Discharges completed within three hours of the discharge order were associated with a 2.0-percent lower length of stay index.
Discharge orders written by 10 a.m. were associated with a 10.2-percent higher length of stay index. Patients discharged by 1 p.m. were associated with a 5.5-percent higher length of stay index.
In terms of ED boarding times, the analysis explored psychiatric and nonpsychiatric ED patients. In the nonpsychiatric population, 33 percent of the variance was explained by the three leading metrics. Discharges completed within three hours of the discharge order correlated with a 24.3-percent increase in the ED boarding times. In contrast, discharge orders written by 10 a.m. correlated with a 9.5-percent decrease in ED boarding times. Patients discharged by 1 p.m. correlated with an 18.1-percent decrease in ED boarding times.
For the psychiatric ED patients, only 0.6 percent of the variance was explained by the leading metrics. Discharges within three hours of the order correlated with an 8.5-percent increase in ED boarding times. Discharge orders written by 10 a.m. correlated with a 2.4-percent increase in boarding. Discharges by 1 p.m. correlated with a 10.1-percent increase in boarding.
When analyzing PACU boarding, only 1 percent of the variance could be explained by the leading metrics. Completing discharges within three hours of the discharge order was associated with a 5.2-percent decrease in PACU boarding times. Discharge orders written by 10 a.m. were associated with a 6.9-percent decrease in PACU boarding times. Patients discharged by 1 p.m. were associated with a 4.0-percent decrease in PACU boarding.
The leading metrics--discharge orders by 10 a.m, discharges within three hours of a discharge order, and discharges by 1 p.m.--had varying correlations with the throughput metrics in this study. In many instances, improvement in these leading metrics was associated with worse throughput outcomes. Based on this, providers may have had a perverse incentive to keep patients longer when they failed to meet the metrics on one day so that they would meet these goals the following day.
Discharges within three hours of the discharge order was the only leading metric found to correlate with lower operational length of stay and LOS index while the others had an opposite effect. However, the variance in these LOS metrics from these leading metrics was minimal.
In contrast, boarding times for the ED nonpsychiatric and PACU patients were significantly impacted by improvement in all three of these throughput metrics, although the variance was drastically different in these two patient populations in that 33 percent of the variance was explained in the ED population while only 1 percent was explained in the PACU setting. ED psychiatric patients were found to have worse boarding times when these leading metrics improved on the inpatient setting.
There are also many other factors that impact stay length, including challenges with placement of patients after discharge, family support, diagnostic delays, etc. However, it was presumed in this study that these types of factors had an equal distribution across the patient population and did not confound the findings in this study.
In conclusion, the leading metrics of discharge orders by 10 a.m., discharges within three hours of a discharge order, and discharges by 1 p.m. were found to have little impact on the inpatient throughput metrics of operational length of stay and LOS index. In most instances, the impact they did have led to an inverse result of longer inpatient stays, which is congruent with the study by Rajkomar, et al. (6) PACU boarding was minimally affected in a positive way by these leading metrics. Only ED boarding of nonpsychiatric patients was found to be positively affected by these metrics. This is similar to findings of a study where more discharges before noon led to ED admissions earlier in the day. (9)
Based on these findings, while improving these leading metrics may not improve a hospital's inpatient length of stay, they likely help in the movement of patients during the day, especially to get them out of the ED and into inpatient beds. Therefore, other leading metrics to help inpatient services improve stay length bear future exploration.
Daniel A. Handel, MD, MBA, MPH, is chief medical officer for the Medical University of South Carolina Medical Center. firstname.lastname@example.org
Matthew Davis, MA, is a data scientist for the Medical University of South Carolina.
Lauren Parnell, MSHCA, ATC, is director of solutions consulting for the Medical University of South Carolina.
Nancy Hendry, RN, MSN, is director of patient logistics for the Medical University of South Carolina Medical Center.
(1.) Hauck K, Zhao X. How dangerous is a day in hospital? a model of adverse events and length of stay for medical inpatients. Medical Care 49(12)106875, December 2011.
(2.) Lim SC, Doshi V, Castasus B, et al. Factors causing delay in discharge of elderly patients in an acute care hospital. Ann Acad Med Singapore 35(1):27-32, January 2006.
(3.) Jaeker JB, Tucker AL. Hurry up and wait: differential impacts on congestion, bottleneck pressure, and predictability on patient length of stay. Harvard Business School Working Paper 13-052, December 2012.
(4.) Kravet 5, Levine R, Rubin H, et al. Discharging patients earlier in the day: a concept worth evaluating. Health Care Manager 26(2)142-6, April/June 2007.
(5.) Powell ES, Khare RK, Venkatesh AK, et al. The relationship between inpatient discharge timing and emergency department boarding. J Emerg Med 42(2)186-96, February 2012.
(6.) Rajkomar A, Valencia V, Novelero M, et al. The association between discharge before noon and length of stay in medical and surgical patients. J of Hosp Med 11(12):859-61, December 2015.
(7.) Wong HJ, Wu RC, Caesar M, et al. Smoothing inpatient discharges decreases emergency department congestion: a system dynamics simulation model. Emerg Med J 27(8):593-8, August 2010.
(8.) Vizient 2016 risk adjustment methodology, (amc.vizientinc.com/docs/5555615580_RiskAdjustmentMethodology2016.pdf). Accessed May 15, 2017.
(9.) Wertheimer B, Jacobs RE, Iturrate E, et al. Discharge before noon: Effect on throughput and sustainability. J Hosp Med 10(10):664-9, October 2015.
With contributions from Anne Wilkinson at the Medical University of South Carolina.
TABLE 1: AGE DISTRIBUTION Under 12 75-older 65-74 55-64 45-54 35-44 25-34 18-24 12-17 TABLE 2: SEX DISTRIBUTION Unknown I Male Female TABLE 3: RACE DISTRIBUTION White/Caucasian Unknown Patient Refused Other Hawaii Native/ Pacific Islander Black/African-American Asian American Indian/ Alaska Native TABLE 4: TOP 20 DIAGNOSIS-RELATED GROUPS Psychoses Vaginal Delivery without Complicating Diagnoses Neonate with Other Significant Problems Normal Newborn Cesarean Section with CC/MCC Seizures without MCC Septicemia or Severe Sepsis without MV>96 hours with MCC Red Blood Cell Disorders without MCC Depressive Neuroses Vaginal Delivery with Complicating Diagnoses Major Joint Replacement or Lower Extremity Reattachment without MCC Alcohol/Drug Abuse or Dependence without Rehab Therapy without MCC Behavioral and Developmental Disorders Esophagitis, Gastroent and Misc. Digestive Disorders without MCC Full-Term Neonate with Major Problems Chemotherapy without Acute Leukemia as Secondary Diagnosis with CC Intracranial Hemorrhage or Cerebral Infarction with MCC Extreme Immaturity or Respiratory Distress Syndrome, Neonate Craniotomy and Endovascular Intracranial Procedures with MCC Other Antepartum Diagnoses with Medical Complications TABLE 5: ACTUAL LENGTH OF STAY PSYCHIATRIC VS. NON-PSYCHIATRIC DIAGNOSES NON-PSYCH 8.12 PSYCH 10.72 Note: Table made from bar graph.
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||FIELD REPORT|
|Author:||Handel, Daniel A.; Davis, Matthew; Parnell, Lauren; Hendry, Nancy|
|Publication:||Physician Leadership Journal|
|Date:||Nov 1, 2017|
|Previous Article:||PULL THE RIGHT LEVERS TO BOOST O.R. VALUE.|
|Next Article:||HOW AN URBAN E.D. IS MAKING CARE BETTER.|