Health-care leaders place significant focus on reducing the average length of stay (ALOS). We examined the relationships among ALOS, cost and clinical outcomes using a neonatal intensive care unit (NICU) simulation model.
A discrete-event NICU simulation model based on the Duke NICU was created. To identify the relationships among ALOS, cost and clinical outcomes, we replaced the standard probability distributions with composite distributions representing the best and worst outcomes published by the National Institutes of Health Neonatal Research Network.
Both average cost per patient and average cost per ⩽28 week patient were lower in the best NICU ($16,400 vs $19,700 and $56,800 vs $76,700, respectively), while LOS remained higher (27 vs 24 days).
Our model demonstrates that reducing LOS does not uniformly reduce hospital resource utilization. These results suggest that health-care leaders should not simply rely on initiatives to reduce LOS without clear line-of-sight on clinical outcomes as well.
Today's health leaders place significant focus on reducing length of stay (LOS). For patients, a shorter LOS means less time away from home and fewer opportunities to develop a hospital-acquired complication. For hospitals, reduced average LOS (ALOS) per global payment should yield a higher per-patient margin. Additionally, some authors have demonstrated a positive correlation between shorter length of stay and better clinical outcomes, lower mortality and fewer readmissions.1, 2, 3 However, others have demonstrated improved outcomes—including decreased mortality—with initiatives that actually lengthen hospital stay,4 while still others have demonstrated in specific cases that a shorter length of stay is associated with increased mortality after discharge.5 Clearly as America transitions from a fee-for-service to a value-based population health system, our effort to find the balance between maximum efficiency and maximum quality represents a major challenge.
Several authors have created models to predict LOS within the NICU based on prenatal data, admission data or outcomes within the first week of life,6, 7, 8, 9 but these models have not captured costs or the impact of variance in outcomes on LOS. Numerous studies have demonstrated correlations between increased LOS and specific morbidities,10, 11 as well as interventions designed to reduce morbidity that correlate with reduced LOS.7, 12, 13 Others have demonstrated that centers with the lowest odds of a prolonged hospital stay also had the highest rates of neonatal mortality.14 In this conflicting milieu, we sought to develop a unifying analysis to model relationships among cost, LOS and neonatal outcomes and definitively answer whether shorter LOS at a unit-level can be uniformly correlated to both better outcomes and lower costs.
Materials And Methods
We first developed a discrete-event simulation model of the Duke University Hospital NICU system using SAS Simulation Studio 14.1 (Cary, NC, USA).15 Discrete-event simulation was selected as an analysis method due to the significant random components and the extremely complicated and intricate mathematical and logical relationships inherent in the NICU system. Furthermore, changes to the state of the NICU system only occur at discrete points in time, where a discrete event might be the arrival of a baby to the unit or the transfer of a baby from the NICU to another hospital. This project sought to address two NICU Management business intelligence needs: (1) the need for an accurate method of modeling patient mix, patient acuity, staffing needs and costs in the present state; and (2) the need to forecast how changes in a unit’s physical structure, staffing, referral patterns, clinical outcomes or patient mix will affect the NICU in a future state.
The full details of this model have previously been published,16, 17 but in brief the functionality of the simulated NICU is defined as follows. Admissions, crashes and deaths can occur at any point during a shift. At the end of each 12-h shift, each patient’s acuity is individually recalculated, patients are transferred out if necessary, patients are moved from critical-care beds to stepdown beds if appropriate and the nurses are reassigned.
The objects, or entities, that flow through the simulated NICU represent patients, and each patient entity has properties, or attributes, that include gestational age, days of life upon arrival and acuity. The number of patients that arrive at the NICU each simulated day is sampled from a probability distribution based on historical data.
The initial model inputs included data on the physical size, structure, nursing practices, patient demographics and patient outcomes across 5 years of experience within the Duke NICU. These data were analyzed, anonymized and combined with published national-level outcomes data18 to generate probability distributions around both admission and morbidity inputs. Using these distributions as inputs to the model, each simulated patient’s entire course on admission to the NICU was predefined, including if and when a crash will occur, shift-to-shift variations in acuity and whether or not the simulated patient will survive to discharge.
To further assess the relationship between ALOS, cost and outcomes, the baseline morbidity probability distributions were replaced with composite distributions to represent the best and worst Neonatal Research Network outcomes.18 The definitions and distributions were based on those of the Neonatal Research Network for necrotizing enterocolitis (NEC, Bell’s stage 2–3), late onset sepsis and intraventricular hemorrhage (IVH, all grades).18 JMP Pro 12 (SAS, Cary, NC, USA)19 was used for statistical analyses.
Representative costs were calculated using previously published per incident costs associated with the neonatal morbidities included in the simulation, marginal effect cost was used for NEC and late onset sepsis while the difference in mean costs between presence and absence of IVH was used.20, 21 These costs were congruent with costs derived from Duke internal data (data not shown). The costs per day were determined by staff acuity coding and appropriate daily costs were allocated to patients at the critical, intensive and stepdown levels. The cost per infant in the simulation was determined and an aggregate was calculated for all infants at the end of the simulation.
As a demonstration of model validity, results averaged from 50 independent replications of the model were compared with the averaged actual data from the Duke NICU for a 5-year period 2008 to 2012 (Table 1). All outcome variables from the simulation model, taken individually, were statistically no different from the actual data. The probability that all seven confidence intervals in Table 1 from the model simultaneously contain their respective true measures is at least 0.65. Note that the results in the third column of Table 1 for the Duke NICU represent those produced by an updated and refined process than was previously published,16 and are based on averaging all values over each of the 5 years separately and then computing a confidence interval for the mean of the five averaged yearly values.
Table 2 reports the outcomes for both the composite best and composite worst performing virtual NICUs. Both average cost per patient and average cost per ⩽28 week gestation patient were lower in the composite best virtual NICU ($16,400 vs $19,700 and $56,800 vs $76,700, respectively), while LOS remained higher in the best performing unit (27 vs 24 days).
Tables 3 and 4 report the paired-t confidence intervals for the difference between the composite best and composite worst outcomes. None of the 95% confidence intervals in Table 3 (taken individually) contain zero, implying that the difference between the composite best and composite worst NICU for each output measure is statistically significant. The probability that all six confidence intervals in Table 3 simultaneously contain their respective true difference is at least 0.70. Similarly, none of the 95% confidence intervals in Table 4 contain zero, implying that the difference in the composite best and composite worst performing virtual NICU is statistically significant for each individual cost measure. The probability that all five confidence intervals in Table 4 simultaneously contain their respective true difference is at least 0.75.
Within neonatology and across medicine, the authors have sought to both model and reduce LOS.22 As health care continues into the era of perpetual reform and cost containment, increasing pressure is being placed on providers, health systems and individual NICUs to cut costs while maintaining or improving quality. The old paradigm of cost reduction focused almost exclusively on either reducing daily costs or reducing LOS, leading to largely one-sided initiatives. This is especially true for the NICU where costs are relatively high and stays are lengthy.23
Previous authors have attempted to model and explain variation in LOS in the NICU.6, 7, 8, 9 These studies highlighted factors that help to explain the differences seen in LOS by describing the limitation of prenatal factors6, 8 and the inclusion and predictive value of later-occurring morbidities.6 These models were limited by the retrospective manner in which they described LOS by attempting to explain the LOS variance in patients who had already had hospitalizations. Our model is unique in that the variance of patient attributes was created de novo using characteristics known to impact LOS in the NICU. The simulated NICU was then retrospectively validated against the unit database for the Duke NICU (Table 1). The model was able to accurately predict common characteristics useful in overall unit functioning as well as the primary outcome measure (LOS).
It is important to emphasize that not only did our simulation demonstrate that the best performing virtual NICU had an overall longer ALOS (27 vs 24 days), longer ALOS for infants ⩽28 weeks (86 vs 66 days) and lower composite cost ($14.1 vs $17.8 million), but that these outcomes were largely driven by morbidity and mortality. For example, the best performing unit had markedly fewer patients IVH and its associated inpatient costs (P<0.001). While calculation of the subsequent lifetime costs associated with the caring for this is beyond the scope of the present paper, the implications of a 59% reduction in IVH should not be overlooked given recent literature questioning the long-term neurodevelopmental impact of even low-grade IVH.24 The same could be said for long-term costs associated with NEC, sepsis, patent ductus arteriosis and retinopathy of prematurity.
While these findings may at first seem counterintuitive, they are in fact consistent with previously reported ‘real world’ clinical data reported by Cotten et al.14 in which they also described an inverse correlation between risk of prolonged hospitalization and mortality. In a large study from the California Quality Care Collaborative, Lee et al.25 found no significant relationship between mortality rate and length of stay.25 However, they noted that the variation between hospitals was significant, suggesting that there was an opportunity to reduce the LOS across different NICUs.25 In considering these results, it is possible that the clinical heterogeneity of their state-wide NICUs obscured a relationship between mortality rate and LOS. Reduction in clinical heterogeneity within the Neonatal Research Consortium may help to account for their findings identifying relationships between clinical morbidities, mortality and LOS.6, 14 In the present study in which all inter-institutional variance has been eliminated, the factor most likely associated with the lengthening of the ALOS observed in the better performing units was their decrease in infant mortality.
As previously reported, the presence of neonatal morbidities is directly related to a lengthy LOS.6, 7, 8, 9 Extremely premature infants (post menstrual age⩽28 weeks) not only experience particularly high costs of care, but also are the infants most likely to experience significant morbidity in the neonatal period.20, 26 Our simulation model demonstrated that with respect to this population, ALOS was higher in the virtual unit with the lowest cost (a difference of 21 days on average, see Table 3) likely demonstrating a counterbalancing affect between reducing mortality and also reducing morbidity. Taken together, our observations suggest that by targeting efforts to reduce specific neonatal morbidities like NEC and IVH, NICUs may be able to decrease their overall costs of this very expensive population despite an increase in LOS.
Our study has several limitations, including the retrospective nature of the data used to build our model. As a simulation, the model relies heavily on the use of pre-existing data to develop the necessary algorithms to predict the outcome measures. As a result, there exists an inherent bias with respect to deaths, NEC, bronchopulmonary dysplasia and any other postnatal age dependent morbidity. These entities may become co-variate and biased (e.g., early death tends to preclude both NEC and bronchopulmonary dysplasia). In the present model, these variables were treated as independent factors based on a retrospective statistical analyses of clinical outcome data.17
Our estimates of costs are limited by both the global nature in which we had to apply per-diem costs based on acuity and the uniform nature in which we applied incremental morbidity-related costs published in the literature. While we sought to estimate total costs as accurately as we could, given the potential variation in cost among units and varying methodologies used to estimate incremental morbidity-related costs (especially when morbidities are rarely truly independent variables), it is possible that we have either over- or underestimated actual cost of care. However, a major benefit of the discrete-event simulation modeling approach used in this project is that with modification of certain critical variables unique to an individual NICU—number of critical care and stepdown beds, number of nurses, transfer pattern, admission distributions and costs—this model is potentially transferrable to other units. Although beyond the scope of this study, to the extent that basic costs (personnel, supplies, procedures, etc.) vary widely between units, it may be possible to construct future models that would focus on resource utilization rather than costs per se. However, to the extent that hospitals may vary widely as to their use of resources for even common diseases, it remains to be seen whether a general description of resource utilization will be helpful when applied to a local health-care system.
Our simulation model is, to our knowledge, the first prospective model system that has been validated to actual NICU performances.14, 18 Its ability to adequately model variations in LOS by using probability distributions for neonatal morbidity and admission inputs represents a unique tool to both drive unit innovation and understand complex relationships among costs. Contrary to current belief, our model demonstrates that shorter LOS is not uniformly associated with reduced hospital resource utilization. These results suggest that hospital leaders and government agencies should strongly consider adding high-fidelity modeling to identify cost-saving opportunities rather than simply relying on initiatives to reduce LOS.
Osnabrugge RL, Speir AM, Head SJ, Jones PG, Ailawadi G, Fonner CE et al. Cost, quality, and value in coronary artery bypass grafting. J Thorac Cardiovasc Surg 2014; 148 (6): 2729–2735.
Kaboli PJ, Go JT, Hockenberry J, Glasgow JM, Johnson SR, Rosenthal GE et al. Associations between reduced hospital length of stay and 30- day readmission rate and mortality: 14-year experience in 129 Veterans Affairs hospitals. Ann Intern Med 2012; 157 (12): 837–845.
Stambough JB, Nunley RM, Curry MC, Steger-May K, Clohisy JC . Rapid recovery protocols for primary total hip arthroplasty can safely reduce length of stay without increasing readmissions. J Arthroplasty 2015; 30: 521–526.
Tadros RO, Faries PL, Malik R, Vouyouka AG, Ting W, Dunn A et al. The effect of a hospitalist comanagement service on vascular surgery inpatients. J Vasc Surg 2015; 61: 1550–1555.
Nordstrom P, Gustafson Y, Michaelsson K, Nordstrom A . Length of hospital stay after hip fracture and short term risk of death after discharge: a total cohort study in Sweden. BMJ (Clin Res Ed) 2015; 350: h696.
Hintz SR, Bann CM, Ambalavanan N, Cotten CM, Das A, Higgins RD . Predicting time to hospital discharge for extremely preterm infants. Pediatrics 2010; 125 (1): e146–e154.
Bender G . Neonatal intensive care unit: predictive models for length of stay. J Perinatol. 2013; 33 (2): 147–153.
Lee HC, Bennett MV, Schulman J, Gould JB . Accounting for variation in length of NICU stay for extremely low birth weight infants. J Perinatol 2013; 33 (11): 872–876.
Pepler P, Uys D, Nel D . Predicting mortality and length-of-stay for neonatal admissions to private hospital neonatal intensive care units: a Southern African retrospective study. Afr Health Sci 2012; 12 (2): 166–173.
Bisquera JA, Cooper TR, Berseth CL . Impact of necrotizing enterocolitis on length of stay and hospital charges in very low birth weight infants. Pediatrics 2002; 109 (3): 423–428.
Stoll BJ, Hansen N, Fanaroff AA, Wright LL, Carlo WA, Ehrenkranz RA et al. Late-onset sepsis in very low birth weight neonates: the experience of the NICHD Neonatal Research Network. Pediatrics 2002; 110 (2 Part 1): 285–291.
Rogowski JA, Horbar JD, Plsek PE, Baker LS, Deterding J, Edwards WH et al. Economic implications of neonatal intensive care unit collaborative quality improvement. Pediatrics 2001; 107 (1): 23–29.
Horbar JD, Rogowski J, Plsek PE, Delmore P, Edwards WH, Hocker J et al. Collaborative quality improvement for neonatal intensive care. NIC/Q Project Investigators of the Vermont Oxford Network. Pediatrics 2001; 107 (1): 14–22.
Cotten CM, Oh W, McDonald S, Carlo W, Fanaroff AA, Duara S et al. Prolonged hospital stay for extremely premature infants: risk factors, center differences, and the impact of mortality on selecting a best-performing center. J Perinatol 2005; 25 (10): 650–655.
SAS Institute Inc SAS Simulation Studio 14.1: User’s Guide. SAS Institute: Cary, NC, 2015.
DeRienzo C, Shaw RJ, Meanor P, Lada E, Ferranti MJ, Tanaka D . A simulation tool to support and predict hospital and clinic staffing. Health Informatics J 2016 (e-pub ahead of print 29 February 2016; doi:10.1177/1460458216628314).
DeRienzo C, Tanaka D, Lada E, Meanor P Creating a SimNICU: Using SAS Simulation Studio to Model Staffing Needs in Clinical Environments. Proceedings of the SAS Global Conference, Washington, DC, 2014.
Stoll BJ, Hansen NI, Bell EF, Shankaran S, Laptook AR, Walsh MC et al. Neonatal outcomes of extremely preterm infants from the NICHD Neonatal Research Network. Pediatrics 2010; 126 (3): 443–456.
SAS Institute Inc JMP Pro 12: User’s Guide. SAS Institute: Cary, NC, 2015.
Russell RB, Green NS, Steiner CA, Meikle S, Howse JL, Poschman K et al. Cost of hospitalization for preterm and low birth weight infants in the United States. Pediatrics 2007; 120 (1): e1–e9.
Johnson TJ, Patel AL, Jegier BJ, Engstrom JL, Meier PP . Cost of morbidities in very low birth weight infants. J Pediatr 2013; 162 (2): 243–249.
Schulman J . Studying determinants of length of hospital stay. J Perinatol 2006; 26: 243–245.
Levit K, Ryan K, Elixhauser A, Stranges E, Kassed C, Coffey R . HCUP Facts and Figures: Statistics on Hospital-based Care in the United States in 2005. Agency for Healthcare Research and Quality: Rockville, MD, 2007. Available at http://www.hcup-us.ahrq.gov/reports.jsp.
Bolisetty S, Dhawan A, Abdel-Latif M, Bajuk B, Stack J . Intraventricular hemorrhage and neurodevelopmental outcomes in extreme preterm infants. Pediatrics 2014; 133 (1): 55–62.
Lee HC, Bennet MV, Schulman J, Gould JB . Accounting for variation in length of NICU stay for extremely low birth weight infants. J Perinatol 2013; 33 (11): 872–876.
St. John E, Nelson K, Cliver S, Bishnoi R, Goldenberg R . Cost of neonatal care according to gestational age at birth and survival status. Am J Obstet Gyncol 2000; 182 (1): 170–175.
The authors declare no conflict of interest.
About this article
Cite this article
DeRienzo, C., Kohler, J., Lada, E. et al. Demonstrating the relationships of length of stay, cost and clinical outcomes in a simulated NICU. J Perinatol 36, 1128–1131 (2016). https://doi.org/10.1038/jp.2016.128
BMC Pediatrics (2021)