Length of stay in pediatric intensive care unit

Length of stay in pediatric intensive care unit 1.1 Scope of Review The following review of the past work done in the area of intensive care unit (ICU) length of stay is divided into two parts. The first part covers the studies done on the PICU length of stay while the second part delves into the literature of ICU length of stay. 1.2 Studies of Length of Stay in Pediatric Intensive Care Unit Ruttimann Pollack (1996) investigated the relationship of length of pediatric intensive care unit (PICU) stay to severity of illness and other potentially relevant factors available within the first 24 hours after admission. A median and geometric mean length of PICU stay of 2.0 and 1.9 days respectively, and the upper 95th percentile at 12 days were found. To prevent undue influence of outliers, all patients staying longer than 12 days were considered long-stay patients (4.1% of the total sample) and were excluded from the model-building process. In the LOS prediction model, variables found to be significantly associated (p Table 1.1: Log-logistic regression model for length of stay Variable Regression coefficient SE Adjusted LOS ratio 95% CI PRISM score* 0.6386 0.0407 5 1.28 1.25-1.33 10 1.63 1.54-1.74 15 1.80 1.67-1.94 20 1.98 1.82-2.16 25 1.62 1.53-1.72 30 1.29 1.25-1.33 40 1.38 1.33-1.44 50 1.06 1.06-1.07 Primary diagnoses CNS diseases -0.1682 0.0267 0.85 0.80-0.89 Neoplastic diseases 0.2324 0.0579 1.26 1.13-1.41 Drug overdoses -0.1758 0.0383 0.84 0.77-0.90 Inguinal hernia -0.3270 0.1344 0.72 0.55-0.94 Asthma -0.1135 0.0527 0.89 0.80-0.99 Pneumonia 0.2350 0.0475 1.26 1.15-1.39 CNS infections 0.4966 0.0555 1.64 1.47-1.83 Respiratory diseases ÃÆ'- PRISMà ¢Ã¢â€š ¬Ã‚   0.1257 0.0579 1.67 1.49-1.87 Head trauma ÃÆ'- PRISMà ¢Ã¢â€š ¬Ã‚   0.1710 0.0611 1.73 1.53-1.94 Diabetes ÃÆ'- PRISMà ¢Ã¢â€š ¬Ã‚   -0.3332 0.0666 1.23 1.08-1.40 Admission conditions Postoperative 0.1267 0.0243 1.14 1.08-1.19 Inpatient 0.2358 0.0271 1.27 1.20-1.33 Previous ICU admission 0.1562 0.0521 1.17 1.06-1.29 Therapy Mechanical ventilation 0.4900 0.0258 1.63 1.55-1.72 Intercept -0.0191 0.0278 Scale 2.5602 0.0295 Log partial likelihood = -5487.2; global chi-square value = 1601.9; df = 15; p CI, Confidence interval; CNS, Central nervous system *LOS ratios computed relative to PRISM score = 0. à ¢Ã¢â€š ¬Ã‚  LOS ratios computed for an interaction with PRISM score = 6.42 (sample average). Source: Modified from Ruttimann Pollack (1996). In the same study, Ruttimann Pollack (1996) noted the ratio of observed to predicted LOS varied among PICUs from 0.83 to 1.25. The PICU factors associated (p Table 1.2: Effect of PICU characteristics on length of stay Variable Regression coefficient SE Adjusted LOS ratio 95% CI p* Intensivist -0.1208 0.0189 0.89 0.85-0.92 0.0001 Coordination -0.0513 0.0190 0.95 0.92-0.99 0.0071 Residents -0.0586 0.0200 0.94 0.91-0.98 0.0033 ln (PICU/hospital beds) à ¢Ã¢â€š ¬Ã‚   0.0459 0.0170 1.03 1.01-1.06 0.0068 CI, Confidence interval. *2 ÃÆ'- ln (likelihood ratio) test. à ¢Ã¢â€š ¬Ã‚  LOS ratio and 95% CIs computed for and increase of PICU/hospital bed ratio by a factor of 2. Source: Modified from Ruttimann Pollack (1996). Development of a new LOS prediction model was necessary due to the availability of a newly updated pediatric severity-of-illness assessment system, PRISM III-24 (Pediatric risk of mortality, version III, 24-hour assessment). Ruttimann et al. (1998) have then fitted a generalized linear regression model (inverse Gaussian) to the observed LOS data with the log link function. In the new LOS prediction model, variables found to be significantly associated (p Table 1.3: Generalized linear regression model (inverse Gaussian) for length of stay (n = 9558) Variable Length of stay ratio 95% Confidence interval p Valueà ¢Ã¢â€š ¬Ã‚   PRISM III-24 à ¢Ã¢â€š ¬Ã‚ ¡ à ¢Ã¢â€š ¬Ã‚ ¡ 0.0001 (PRISM III-24) °Ã‚ °2 à ¢Ã¢â€š ¬Ã‚ ¡ à ¢Ã¢â€š ¬Ã‚ ¡ 0.0001 Primary diagnoses CNS infections 1.41 1.28-1.56 0.0001 Neoplastic diseases 1.22 1.13-1.31 0.0001 Asthma 0.91 0.85-0.96 0.0045 Pneumonia 1.50 1.40-1.61 0.0001 Drug overdoses 0.74 0.70-0.79 0.0001 CV nonoperative 1.22 1.14-1.32 0.0001 CV operative 0.89 0.83-0.95 0.0006 Diabetes 0.74 0.67-0.81 0.0001 Admission specifications Postoperative 0.92 0.88-0.96 0.0004 Inpatient 1.17 1.13-1.22 0.0001 Previous ICU admission 1.26 1.15-1.38 0.0001 Therapy Mechanical ventilation 1.68 1.60-1.77 0.0001 Model intercept ( ± SEM) = 1.423  ± 0.021 days CNS, Central nervous system; CV, cardiovascular system.  °Effect of the variable after adjusting for the effects of all other variables in the model. à ¢Ã¢â€š ¬Ã‚  Log-likelihood ratio compared with the chi-squared distribution with 1 degree of freedom. à ¢Ã¢â€š ¬Ã‚ ¡See Fig.2 (pg 82, Ruttimann et al. 1998). Model fit: Scaled deviance = 9558 (chi-square with 9543 degrees of freedom, p >0.45). Observed versus predicted length of stay, mean ( ± SEM) in: training sample (n = 9,558): 2.351( ± 0.032) versus 2.360( ± 0.011), p >0.64; test sample (n = 1,100): 2.461( ± 0.069) versus 2.419( ± 0.035), p >0.49. Source: Modified from Ruttimann et al. (1998). Ruttimann et al. (1998) have also assessed the PICU efficiency with the new LOS prediction model and validation of the assessment by an efficiency measure based on daily use of intensive care unit-specific therapies (based on the criterion whether on each day a patient used at least one therapy that is best delivered in the ICU). PICU efficiency was computed as either the ratio of the observed efficient days or the days accounted for by the predictor variables to the total care days, and the agreement was assessed by Spearmans rank correlation analysis. PICU efficiency comparisons for both the predictor-based and therapy-based methods are nearly equivalent. Ruttimann and colleagues (1998) acknowledged the advantage of predictor-based efficiency as it can be computed from admission day data only. It was of researchers utmost interest to study the extended LOSs as well. Long-stay patients (LSPs) in the PICU were later being examined by Marcin et al. (2001). As explained previously, LSPs were defined as patients having a length of stay greater than 95th percentile (>12 days). In the study, the clinical profiles and relative resource use of LSPs were determined and a prediction model was developed to identify LSPs for early quality and cost saving interventions. To create a predictive algorithm, logistic regression analysis was used to determine clinical characteristics, available within the first 24 hours after admission that were associated with LSPs. Marcin and colleagues (2001) noted that, Long-stay patients in the PICU consume a disproportionate amount of health care resources and have higher mortality rates than short-stay patients. Multivariate analysis of the study identified predictive factors of long-stay as: age Table 1.4: Significant independent variables from the logistic regression analysis Variable Odds Ratio 95% CI p Value Age 1.77 1.42-2.20 Previous ICU admission 2.18 1.52-3.11 Emergency admission 1.67 1.28-2.19 CPR before admission 0.59 0.37-0.96 0.032 Admitted from another ICU or IMU 2.28 1.13-4.58 0.020 Chronic TPN 3.09 1.39-6.92 0.006 Chronic tracheostomy 2.23 1.41-3.52 0.001 Pneumonia 2.73 2.03-3.68 Other respiratory disorder 2.33 1.64-3.32 Acquired cardiac disease 3.07 2.01-4.67 Having never been discharged from hospital 2.27 1.12-4.59 0.020 Ventilator 4.59 3.60-5.86 Intracranial catheter 2.78 1.76-4.41 PRISM III-24 score between 10 and 33 2.99 2.35-3.81 CI, confidence interval; ICU, intensive care unit; CPR, Cardiopulmonary resuscitation; IMU, intermediate care unit; TPN, total parenteral nutrition; PRISM, Pediatric Risk of Mortality. Source: Modified from Marcin et al. (2001). In a case study carried out by Kapadia et al. (2000) in a childrens hospital in the Texas Medical Center in Houston, discrete time Markov processes was applied to study the course of stay in a PICU as the patients move back and forth between the severity of illness states. To study the dynamics of the movement of patients in PICU, PRISM scores representing the intensity of illness were utilized. The study modeled the flow of patients as a discrete time Markov process. Rather than describing by a string of services and scores, the course of treatment and length of stay in the intensive care was described as a sequence of Low, Medium and High severity of illness. The resulted Markovian model appeared to fit the data well. The models were expected to provide information of how the current severity of illness is likely to change over time and how long the child is likely to stay in the PICU. The use of a Markovian approach allowed estimation of the time spent by patients in different se verity of illness states during the PICU stay, for the purposes of quality monitoring and resource allocation. 1.2 Studies of Length of Stay in Intensive Care Unit According to Gruenberg et al. (2006), institutional, medical, social and psychological factors collectively affect the length of stay (LOS) in the intensive care unit (ICU). Institutional factors include geographic location, resources, organizational structure, and leadership. In term of medical factors, specific medical interventions, specific clinical laboratory values, and the type and severity of patients illnesses were found to be related to length of stay in the ICU. Social factors such as lack of quality communication between patients families and physicians or other healthcare personnel, and conflict between patients families and hospital staff have resulted in prolonged ICU and hospital stays. Anxiety and depression experienced by a patients family members are psychological characteristics that contribute to inadequate decision making and extended ICU stays. In order to examine the impact of prolonged stay in the intensive care unit (ICU) on resource utilization, Arabi and colleagues (2002) carried out a prospective study to determine the influence of certain factors as possible predictors of prolonged stay in an adult medical/surgical ICU in a tertiary-care teaching hospital. Prolonged ICU stay was defined as length of stay >14 days. The data analyzed included the demographics and the clinical profile of each new admission. Besides, two means were used to assess severity of illness: the Acute Physiology and Chronic Health Evaluation (APACHE) II score (Knaus et al., 1985, as cited in Arabi et al., 2002) and the Simplified Acute Physiology Score (SAPS) II (Le Gall et al., 1993, as cited in Arabi et al., 2002). The study has identified predictors found to be significantly associated with prolonged ICU stay: non-elective admissions, readmissions, respiratory or trauma-related reasons for admission, and first 24-hour evidence of infection, oliguria, coagulopathy, and the need for mechanical ventilation or vasopressor therapy had significant association with prolonged ICU stay (Table 2.5 2.6). It was also found that mean APACHE II and SAPS II were slightly higher in patients with prolonged stay. Arabi et al. (2002) concluded that patients with prolonged ICU stay form a small proportion of ICU patients, yet they consume a significant share of the ICU resources. Nevertheless, the outcome of this group of patients is comparable to that of shorter stay patients. The predictors identified in the study were expected to be used in targeting this group to improve resource utilization and efficiency of ICU care. Table 1.5: Demographic and clinical profile of patients in the study group [all values shown are n (%), except where indicated otherwise] All (n = 947) ICU length of stay p value à ¢Ã¢â‚¬ °Ã‚ ¤ 14 days (n = 843) >14 days (n = 104) Age (years) ¹ 12-44 391 (41.3) 349 (41.4) 42 (40.4) NS 45-64 309 (32.6) 274 (32.5) 35 (33.7) NS à ¢Ã¢â‚¬ °Ã‚ ¥65 247 (26.1) 220 (26.1) 27 (26.0) NS Gender Male 591 (62.4) 518 (61.4) 73 (70.2) NS Female 356 (37.6) 325 (38.6) 31 (29.8) NS Type of admission Elective 169 (17.8) 164 (19.5) 5 (4.8) Non-elective 778 (82.2) 679 (80.5) 99 (95.2) Severity of illness APACHE II score (mean  ± SD) 19  ± 9 19  ± 9 21  ± 8 0.016 SAPS II score (mean  ± SD) 38  ± 20 37  ± 20 43  ± 16 0.003 Tracheostomy 113 (11.9) 52 (6.2) 61 (58.7) ICU mortality 193 (20.4) 173 (20.5) 20 (19.2) NS NS, not significant.  ¹Because of rounding, some of the percentages may not add up to 100% exactly. Source: Modified from Arabi et al. (2002). Table 1.6: Possible predictors for prolonged stay and the associated odds ratio No. of patients (%) ORs for prolonged stay p value (n = 947) OR 95% CI Non-elective admission 778 (82.8) 4.7 1.9-11.7 Readmission 79 (8.3) 2.1 1.1-3.8 0.02 Main reason for admission Surgical Trauma 171 (18.1) 2.1 1.4-3.4 Non-trauma surgical 231 (24.4) 0.3 0.1-0.5 Medical Cardiovascular 212 (22.4) 1.0 0.6-1.6 NS Respiratory 159 (16.8) 2.2 1.4-3.6 Neurologic 36 (3.8) 0.5 0.1-2.0 NS Other 138 (14.6) 0.51 0.25-1.05 NS First 24-hour data Coagulopathy 345 (36.4) 1.5 1.0-2.3 0.05