Objective. It is standard practice to defer discharge of premature infants until they have achieved a set number of days without experiencing apnea. The duration of this period, however, is highly variable across institutions, and there is scant literature on its effectiveness or value-for-money. Our objective was to establish the economic impact of varying durations of predischarge observation for apnea of prematurity.
Methods. Using computer simulation, we compared the alternatives of hospital monitoring for 1 to 10 days, after apparent cessation of apnea, with no monitoring and with the next longest period of monitoring. The daily probability of apnea requiring stimulation after a given number of apnea-free days was obtained from chart review of 216 infants, beginning on the day they attained both full feeds and temperature stability in an open crib. Baseline rates of survival or impairment, utilities for calculation of quality-adjusted life years (QALYs), outcomes for respiratory arrest at home, and long-run costs for neurodevelopmental impairment were derived from the literature. Hospital expenditures were obtained from itemized billing records for infants on each of the final 10 days of hospitalization and converted to costs using Medicare cost-to-charge ratios. Costs are reported in 2000 US dollars.
Results. For infants born at 24 to 26 weeks’ gestation, each additional day of monitoring cost from $41000 per QALY saved for the first day to >$130000 per additional QALY gained for the tenth day. Cost-effectiveness was poorer for infants who were born at gestational ages >30 weeks. Results were sensitive to the proportion of charted apneas requiring stimulation that would actually progress, without intervention, to respiratory arrest.
Conclusions. In this model, the cost-effectiveness of predischarge monitoring for apnea of prematurity declined significantly as the duration of monitoring was increased. Consideration should be given to alternative uses for resources in formulating neonatal discharge guidelines.
- resource utilization
- decision analytic modeling
- cost analysis
- apnea of prematurity
A significant proportion of neonatal care for premature infants occurs during the prolonged period after acute illness but before the establishment of adequate physiologic stability that would permit discharge home. This instability arises from 4 sources: apnea of prematurity, temperature dysregulation, inadequate oral feeding skills or growth, and residual chronic lung disease. It is likely that any interventions that affect these relatively mild derangements will have disproportionate impact on overall neonatal lengths of stay and costs. Similar high-volume, low-acuity conditions have been shown previously to be important determinants of total neonatal expenditures.1
Apnea of prematurity is almost universal in infants before approximately 28 weeks’ postmenstrual age, then gradually becomes less frequent with maturation of the central nervous system through 35 to 36 weeks’ postmenstrual age, although group differences in the incidence of cardiorespiratory events persist to approximately 43 weeks’ postconceptional age.2 Apneic episodes may range in severity from those that resolve spontaneously, through episodes that require stimulation of the infant to reestablish respiratory effort, to severe events requiring artificial ventilation or resuscitation. The disorder is treated with mechanical ventilation at its most acute stage and with medication. During the marginal period before discharge, however, episodes usually fall at the less severe end of the spectrum, and infants are simply monitored off medication using a device that alarms when the heart or respiratory rates fall below a defined threshold. This monitoring continues until the infant is deemed “mature,” that is, has remained asymptomatic for between 3 and 10 days, with the exact duration varying significantly between institutions.3 At this point, it is believed that the likelihood of recurrent apnea is exceedingly remote, and the infant is discharged. Because of the prolonged duration of monitoring, this is often the only reason that the infant must remain in the hospital and thus may be an important determinant of costs related to length of stay.
There are no studies that establish either the efficacy or the cost-effectiveness of the various durations of monitoring. This is presumably because adverse outcomes are rare, and the length of the monitoring period in most institutions is very conservative; direct investigation therefore would require very large cohorts and a major change in practice and might raise ethical issues of randomizing infants to lesser treatment. However, by making the assumptions that the rate of resolution of apnea after discharge is the same as that observed in the hospital and that non–self-resolving apneas at home would lead to respiratory arrest, the process can be modeled using decision-analytic methods.
Study Design and Framing
The study was an incremental cost-effectiveness analysis using decision-analytic modeling. In this approach, data from both primary databases and secondary sources, such as publications, are combined using computer software and are used to simulate costs and outcomes of a particular therapeutic intervention.
The analysis targeted infants at highest risk for apnea of prematurity, specifically those with gestational ages at birth of 24 to 34 weeks by best obstetric estimate. These infants were assumed to have received other aspects of care typical for hospitalized neonates in North American intermediate- and high-risk neonatal intensive care units (NICUs). As the decision for discharge on the basis of apnea of prematurity would apply only when other markers of developmental immaturity had resolved, only infants who had already achieved full oral feeds, who no longer required an incubator for external warming, and who were not receiving drug therapy for apnea were included in the analysis.
Physicians generally choose to discharge infants after a period of observation of 1 to 10 consecutive days without recorded apnea or bradycardia.3 The analysis compared each of these periods of monitoring with the next longest period of monitoring. A 0-day monitoring alternative was also included, as infants are occasionally discharged with active apnea of prematurity on cardiorespiratory monitors, for which the effectiveness in preventing death outside hospital has not been established.
To facilitate comparison with other health-related social programs, we used a societal perspective in deciding which costs and effects to measure. In this comprehensive viewpoint, all relevant costs and outcomes were considered, regardless of the parties to whom they accrued.4 Similarly, because the outcomes of neonatal respiratory arrest include chronic health states with long-term costs, a lifetime horizon was taken, in which all effects and costs of the alternative approaches to monitoring in the neonatal period were modeled, regardless of when they occurred.
Individuals usually prefer to receive benefits earlier and to defer costs so that resources can be put to other uses in the interim. For taking this time preference into account, future costs and effects are typically weighted less heavily through a “discount rate.” In our study, costs and effects for the base case were discounted at 3% per year, and this value was varied between 0% and 5% in sensitivity analyses.4
Events in the neonatal period were described with a state-transition decision-analytic model5 using the DATA 3.5.5 for Healthcare software package (TreeAge Software, Williamstown, MA). The structure of the decision-analytic model is shown in Fig 1 . The model is summarized below. Full details of structure and inputs are available in a technical report from the authors.6
For each of the alternative lengths of monitoring (1–10 days) shown in Fig 1, infants begin their course in the hospital. For the no-intervention (0 days of monitoring) comparator, infants begin their course at home. On any given day, infants can remain asymptomatic or experience apnea. If they remain asymptomatic, then they progress to the next “asymptomatic day.” After 10 asymptomatic days, either in the hospital or at home, infants are assumed to be free of apnea and are denoted to have “no sequelae from apnea.” They may still have neurodevelopmental sequelae related to other neonatal risk factors (most prominently gestational age), independent of apnea.
Apneas that occur in the hospital are treated with stimulation of the infant by gentle patting or in more serious instances by bag-mask ventilation. Most hospitalized infants with apnea that is not self-resolving have no obvious adverse consequences on quality-adjusted life expectancy, because they receive immediate stimulation or, rarely, resuscitation and do not experience the effects of hypoxia. Thus, all infants who experience apnea in the hospital are modeled as simply restarting the apnea watch, without any increased risk for neurodevelopmental sequelae or death. In contrast, some proportion of apneas experienced at home would not resolve spontaneously. It is assumed that, without intervention, these would result in respiratory arrest. Because resuscitation is less successful and more delayed out of the hospital, there is a probability of both death and neurodevelopmental compromise, in addition to the gestation-specific risks.
Face validity of the model was assessed in several presentations to clinicians during its development. Technical accuracy was verified by using inputs for which results were obvious and predictable.7 The modeling exercise was undertaken because outcome data do not exist for varying duration of predischarge monitoring for apnea of prematurity; thus, predictive validity could not be assessed.
Sources for Costs
Daily costs for hospitalized infants were taken from patient-specific itemized charge records for 628 infants who were at 24 to 34 weeks’ gestation and were eventually discharged directly home from the Brigham and Women’s Hospital NICU in 1990 and 1991. Charges were converted to costs using cost center–specific Medicare ratios of costs to charges for the Brigham and Women’s Hospital for 1991. Other cost variables are given in Table 1 and summarized below.
Professional costs for routine neonatal care of infants with minimal illness acuity were inferred from a neonatal labor time and motion study performed by the authors.8 Time was converted to costs using typical clinician salaries for the institution.
Costs for resuscitation were obtained from the clinician fee schedules at Children’s Hospital, Boston, in 1998. The resuscitation protocol was derived from typical clinical practice for unexplained infant respiratory arrest at home. Fees were adjusted by the average departmental collection fraction to reflect actual reimbursement. Costs of emergency medical response were obtained from the fee schedule of a nationally represented ambulance service.
Long-term direct medical costs for infants with neurodevelopmental sequelae were obtained from the cerebral palsy subgroup of an extensive cost of illness study of 18 birth defects.9 This study also included costs of educational and developmental interventions in the long term.
Expenditures such as parental travel costs and child care in the neonatal period were not included, as accurate information on these is not available from secondary sources. However, because these would be differential between the strategies for at most 10 days, the total sum is likely to be small relative to high daily hospital costs.
It was assumed that parents had returned to work in the predischarge period and that they took leave from work during the initial days with the infant at home. National average nonadjusted hourly earnings for production workers were applied to the expected number of weekdays spent at home by 1 parent.10
Costs were expressed in 2000 US dollars. When appropriate, currencies were converted between time periods using the medical care component of the Consumer Price Index.11
Sources for Probabilities and Effects
Input values for probabilities and effects, along with values for sensitivity analyses, are shown in Table 2 . Baseline mortality and morbidity outcomes for premature infants were derived from the literature12–15 and from local data collected at the Brigham and Women’s Hospital in 1996 and 1997 for 565 extremely low birth weight infants (A. Stark, Harvard Medical School, personal communication). Outcomes for infants who experience apnea at home were derived from published studies on out-of-hospital pediatric cardiorespiratory arrest.16–18
Effectiveness outcomes were expressed as both life years and quality-adjusted life years (QALYs) gained. In the latter approach, a patient’s preference, or utility, for a health state is multiplied by the length of time spent in that state to give a QALY. Utilities for calculation of QALYs were obtained from published studies of parents’ assessments of the health status of children born prematurely and control infants, based on the Health Utilities Index.13,14
No studies in the literature measure the probability of developing apnea after a given number of days without symptoms. Values for the probability of remaining asymptomatic between 1 day and the next were derived from a sample of 216 infants who were cared for in the Brigham and Women’s Hospital NICU between 1993 and 1996. Characteristics of these infants are given in Table 3 . Standard practice in the NICU at that time included continuous cardiorespiratory monitoring of all infants throughout the admission. Apnea alarms occurred after 20 seconds of absence of respiratory effort, whereas bradycardia alarms typically were set for heart rates below 80 beats per minute. Infants were typically discharged when they had attained full feeds with adequate growth and temperature regulation and had had no cardiorespiratory events at rest for 5 days. In our study, the daily number of apneas that required stimulation was abstracted retrospectively from patient charts from the day these infants attained both full feeds and temperature stability in an open crib up to discharge. Results were stratified by gestational age, because there is some evidence of a slower resolution of apnea at lower gestations.19 The probabilities were then smoothed using logistic regression to eliminate the effects of sampling error. Figure 2 shows the smoothed curves for the 3 gestational age ranges.
The reliability and validity of nurses’ decisions to record apnea as requiring stimulation have not been studied. The proportion of such apneas that would result in respiratory arrest at home if no intervention occurred is therefore unknown. However, it is very likely that any such readmissions or deaths within a few days of discharge would come to the attention of clinicians in the Division of Newborn Medicine. The database from which the transition probabilities were derived was therefore used as a sample of the population of infants at risk for such arrest. It was assumed that no home arrests had been observed among these 216 infants. A Poisson approximation to the binomial distribution was used to generate the probability that observing no deaths in the group was attributable entirely to chance. The proportion of apnea that led to arrest and resulted in an α error of 0.05 was selected as the baseline value for analyses.
Analysis and Statistical Considerations
The decision-analytic model was used to simulate baseline costs and effects for a cohort of infants in each of 3 gestational age categories: 24 to 26 weeks, 27 to 29 weeks, and 30 to 34 weeks. Incremental cost-effectiveness ratios were calculated as the additional cost of a period of monitoring, divided by the additional effectiveness, compared with the next longest period of monitoring.
For assessing the uncertainty associated with the results, 2 forms of analyses were then performed. In deterministic sensitivity analysis, values for input variables were varied through their plausible ranges (given in Tables 1 and 2), and the changes in the cost-effectiveness ratios were recorded. In probabilistic sensitivity analysis, random uncertainty associated with sampling of any variables that had probability distributions was assessed; this included, for example, variables such as hospital costs or apnea rates that were obtained from cohorts of real patients. The procedure involved randomly sampling from the distributions of each relevant input variable 1000 times and repeating the simulation each time to generate 1000 incremental cost-effectiveness ratios for each period of monitoring. The 975th and 25th largest cost-effectiveness ratios were selected from the ordered list to construct 95% confidence intervals.
Baseline results are presented in Table 4 and Fig 3 . As shown, the mean incremental cost-effectiveness values range from approximately $19000 to $41000 per additional QALY for 1 day of observation up to $131000 to $355000 per QALY for 10 days of observation. Although the 24- to 26-week and 27- to 29-week gestation cohorts were similar, there was a tendency toward poorer cost-effectiveness of monitoring for infants who were born at more mature gestational ages, especially for the longer monitoring alternatives. Thus, a 7-day period of monitoring cost $179000 per additional QALY for the 30- to 34-week cohort and only $97000 for the 24- to 26-week cohort. Results expressed as dollars per life year were similar to those for dollars per QALY.
In sensitivity analyses, results were very stable (change in incremental cost-effectiveness <10%) to changes in the following model assumptions: worse literature-derived outcomes for out-of-hospital respiratory arrest, lower utilities for the in-hospital state, higher or lower hourly wage, higher or lower physician salary, or worse baseline outcomes of prematurity. Results were moderately stable (change in incremental cost-effectiveness 10%–30%) to better outcome of out-of-hospital respiratory arrest, or changes in acute neonatal hospital costs. There was a substantial improvement (change in incremental cost-effectiveness 30%–50%) in cost-effectiveness ratios when it was assumed that a higher proportion of charted apneas would lead to respiratory arrest if no intervention occurred (ie, when it is assumed that 1 home arrest occurred in the 216-infant sample without clinicians becoming aware of it). Similar differences were seen when the discount rate for costs and effects was varied between 0% and 5%. In addition to these variables, the results were sensitive to the gestational age category as noted above, as a result of the differing baseline incidence of apnea.
As noted above, there was a trend toward poorer cost-effectiveness for infants in the more mature 30- to 34-week gestation cohort. Moreover, this group of infants is much more numerous, composing 80% of the 173000 births below 34 weeks in the United States (J. Martin, National Center for Health Statistics, personal communication, October 1997). If total NICU beds are constrained, then a rational economic policy might dictate putting more resources into monitoring less mature rather than more mature infants. Table 5 shows the implications of such differential monitoring for a hypothetical cohort of 1000 infants. Compared with the typical strategy of monitoring all infants for 5 days, a policy of monitoring infants of 30 to 34 weeks’ gestation for 4 days and those below 30 weeks’ gestation cohorts for 7 days would consume 453 fewer bed days, cost $621000 less, and generate 21 more QALYs. The differential policy is therefore dominant in that it both reduces costs and improves outcomes.
This study is the first to explore systematically the economic implications of the current practice of monitoring for resolution of apnea of prematurity for a fixed number of days before discharge home. The results show that the incremental cost-effectiveness ratios for monitoring infants of 24 to 26 weeks’ gestation for only 1 asymptomatic day before discharge exceeded $41000 per QALY, and addition of a second day further worsened the cost-effectiveness to >$69000 per QALY. The cost-effectiveness exceeded $100000 per QALY by 8 days in infants <30 weeks’ gestation and by 5 days for those >30 weeks. Thus, there was both a steep deterioration in value-for-money as the length of monitoring increased and a worsening cost-effectiveness for infants who were born at higher gestational ages.
The model had several strengths not typically seen in pediatric modeling studies. These included the use of actual patient data for both costs and probability of apnea, adherence to a marginal approach for both costs and effects, and the use of simulation modeling to generate 95% confidence intervals.
Before consideration of the implications of the results, however, it is also important to understand the assumptions and limitations of the simulation approach. The model was based on 3 main structural assumptions and a number of assumptions regarding the individual inputs. The first structural assumption was that the rate of apnea in the hospital is the same as it would be if the infants were at home. Although this assumption cannot be tested formally, there is no biological rationale against it; in the few days before discharge, the observed infants did not receive any confounding interventions, such as drug therapy or respiratory support.
The second main assumption was that the outcome of respiratory arrest at home is the same as it would be for other causes unrelated to prematurity. In this case, because the literature outcomes are already very poor, using a worse outcome in sensitivity analyses had very little effect on the results, whereas using more optimistic outcomes tended to worsen the cost-effectiveness by up to 30%.
The final main assumption related to the rate of progression of apnea to full respiratory arrest. As noted in “Methods,” this rate is not observable in the hospitalized sample, as apnea that seems to be persistent (“apnea requiring stimulation”) always results in intervention, specifically gentle tactile stimulation. The rate was therefore inferred from the sample of infants who generated the apnea probabilities, by assuming that the observed 0 event rate was consistent with an α sampling error of 0.05. Although it is highly unlikely that any of these infants experienced a respiratory arrest at home without the involved clinicians being informed, it is conceivable that repeated samples of n = 216 might contain a case of an infant who died from arrest at home. In sensitivity analysis, there was a substantial improvement in cost-effectiveness ratios when the rate was assumed to be higher. However, even under this unlikely scenario, the cost-effectiveness of monitoring exceeded $50000 per QALY by 5 days in all 3 gestational age cohorts.
With regard to specific choices for input variables, sensitivity analyses showed the results to be very stable or moderately stable to changes in hospital costs, patient preferences for remaining in the hospital, physician and parent salaries, and baseline prematurity outcomes. However, there was substantial change in the cost-effectiveness ratios when the model assumed a 0 discount rate. This is not specific to apnea of prematurity: most interventions in which resources are used early to change later outcomes—as is the case for almost every neonatal therapy—will be more attractive when the later outcomes are not discounted. Moreover, the absence of a time preference is not consistent with empirical studies. Of note, varying the discount rate to 5%, as has been suggested in some pharmacoeconomic guidelines, resulted in a precipitous worsening of the cost-effectiveness.
Finally, it is possible that there is some selection bias in the observational component of the study. Infants who are discharged directly home from the tertiary care hospital rather than being transferred to a community hospital first may be sicker and thus more likely to have apnea. This would result in a higher rate of apnea, which would in turn lead to better cost-effectiveness ratios. Thus, if there were selection bias, then the results would underestimate rather than overestimate the true cost per QALY saved.
Another limitation is that this was a simulation study. Unfortunately, experimental or observational studies of appropriateness of in-hospital monitoring are exceedingly difficult to perform for apnea of prematurity for several reasons. These include the rarity of adverse events, entrenched clinical dogma regarding the monitoring period, and both ethical and medical/legal resistance to changing current practice to a less conservative approach. This resistance persists despite substantial equipoise when the literature is examined objectively. In the absence of more rigorous study designs for direct economic evaluation, the only alternative is to attempt to synthesize the available data in as transparent a manner as possible and generate estimates of cost-effectiveness that can be refined later.
It is apparent, then, that the exact values of the cost-effectiveness ratios must be taken with caution. However, the more general result—that each additional day of monitoring results in steeply increasing incremental cost-effectiveness ratios—was robust. There are important policy implications to these findings. Although methodological differences cause potential problems in comparing cost-effectiveness ratios between studies,20 this range of cost-effectiveness is at or beyond the upper range for other accepted health interventions,21 particularly those in infant populations, such as hepatitis B immunization,22,23 neonatal intensive care,24 or its better-studied components, such as surfactant25–29 or erythropoietin30 therapy. More important, the differences in cost-effectiveness across gestational age cohorts suggests that monitoring lower gestation infants for a longer period and higher gestation infants for a shorter period may result in better outcomes and decreased costs compared with the status quo. Such cost-saving policies are rare in neonatology, surfactant being the single well-studied example.25–29
Although the results of modeling studies such as this one cannot guide a change in policy on their own, they do point to the high priority for additional research into high-volume, low-acuity care. Prospectively planned economic evaluation should be an integral component of any such research.
We thank the members of the Division of Newborn Medicine, Harvard Medical School, for helpful comments in reviewing the structure of the analytic model. Chart abstraction of rates of apnea in hospital was performed by summer student Emma Jones.
- ↵Zupancic JA, Richardson DK. Characterization of the triage process in neonatal intensive care. Pediatrics.1998;102 :1432– 1436
- ↵Darnall RA, Kattwinkel J, Nattie C, Robinson M. Margin of safety for discharge after apnea in preterm infants. Pediatrics.1997;100 :795– 801
- ↵Gold MR, Siegel JE, Russell LB, Weinstein MC, eds. Cost-Effectiveness in Health and Medicine. New York, NY: Oxford University Press; 1996
- ↵Sonnenberg FA, Beck JR. Markov models in medical decision making: a practical guide. Med Decis Making.1993;13 :322– 338
- ↵Zupancic J. Economic Evaluation of Neonatal Intensive Care. Boston, MA: Harvard University; 2001
- ↵Zupancic JAF, Richardson DK. Prediction of neonatal personnel time inputs from clinical variables. A time and motion study. J Perinatol. In press
- ↵Waitzman NJ, Scheffler RM, Romano PS. The Cost of Birth Defects. Estimates of the Value of Prevention. Lanham, MD: University Press of America; 1996
- ↵United States Bureau of Labor Statistics. National Employment, Hours and Earnings. Washington, DC: Bureau of Labor Statistics; 1998
- ↵United States Bureau of Labor Statistics. Consumer Price Index–All Urban Consumers–Medical Care (Series ID MUUS0000SA5). Washington, DC: Bureau of Labor Statistics; 2000
- ↵Hack M, Wright LL, Shankaran S, et al. Very-low-birth-weight outcomes of the National Institute of Child Health and Human Development Neonatal Network, November 1989 to October 1990. Am J Obstet Gynecol.1995;172(2 Pt 1) :457– 464
- ↵Eichenwald EC, Aina A, Stark AR. Apnea frequently persists beyond term gestation in infants delivered at 24 to 28 weeks. Pediatrics.1997;100(3 Pt 1) :354– 359
- ↵Krahn M, Detsky AS. Should Canada and the United States universally vaccinate infants against hepatitis B? A cost-effectiveness analysis. Med Decis Making.1993;13 :4– 20
- ↵Mugford M, Piercy J, Chalmers I. Cost implications of different approaches to the prevention of respiratory distress syndrome. Arch Dis Child.1991;66(7 Spec No) :757– 764
- Maniscalco WM, Kendig JW, Shapiro DL. Surfactant replacement therapy: impact on hospital charges for premature infants with respiratory distress syndrome. Pediatrics.1989;83 :1– 6
- ↵Tubman TR, Halliday HL, Normand C. Cost of surfactant replacement treatment for severe neonatal respiratory distress syndrome: a randomised controlled trial. BMJ.1990;301 :842– 845
- Sheikh A, Brogan T. Outcome and cost of open- and closed-chest cardiopulmonary resuscitation in pediatric cardiac arrests. Pediatrics.1994;93 :392– 398
- National Center for Health Statistics. United States Decennial Life Tables for 1989–1991. Vol 1, No 1. Rockville, MD: Public Health Service; February 1998
- Copyright © 2003 by the American Academy of Pediatrics