There are a small number published trials and observational studies that partially describe the BPD distribution of extreme preterm infants and the relative burden of BPD (1-3,11,15, 32, 33). Our study adds a significant expansion of existing evidence by presenting the lifetime burden of BPD based on key patient characteristics. We have shown that the present value of the lifetime expected burden of a BPD patient discharged in 2018 is over $700,000 to the healthcare system. Excluding the high costs incurred in the first year, BPD costs the health system $4,600 per patient per year, making it roughly twice as expensive as type II diabetes annually (29, 30) while the BPD cost is incurred at the very start of a patient’s life. Moreover, the expected QALYs of the average BPD patient is only 42, amounting to an annual expected quality of life score of less than 0.58 (1 indicating perfect health and 0 indicating death). This utility measure may be nearly halved depending on how one accounts for compounding disutility from multiple complications.
This study attributes major preterm complications as part of the burden pathway of BPD, largely to illustrate the association between BPD as an early clinical marker and risk factor for many clinical outcomes resulting from extreme preterm birth. There are several major complications that were not included in our model primarily because we only included complications for which previous literature outlined a differential incidence of complications associated with BPD severity. This model is not meant to reflect the total burden for the average preterm infant, as they may develop complications that are not associated with BPD severity and/or were not captured in the existing BPD literature.
An additional benefit from our simulation model is we now have valuable insight into the scale and shape of the distributions around our prevalence estimates. This is often missing in other model outputs and the vast majority of evidence available to date is based on single sources or time periods, lending no opportunity to apply a range or shape and to characterize the uncertainty around the mean. Figures 2 and 3 present the distributions for the whole population for cost and health utilities obtained from our study, respectively. The cost distribution has an unsurprising right-tail skew indicating a high potential of extreme costs incurred by a portion of patients that pulls the average cost per patient higher. The Appendix I presents several smoothed histograms to show the density of values across all simulants averaged across all runs, as well as the survival curve for all patients that survived to discharge.
The sampling approach used here intentionally applied very little constraint on the sample distribution to allow for our results to be as conservative as possible in attributing high uncertainty to our estimates. Our findings ended up being well aligned to the CNN reported data despite the only use of the CNN data was to assign the rank order of discharge outcomes and to test the bootstrapped samples relative to the CNN reported aggregate mortality confidence range.
There have been well documented attempts at predicting BPD outcome based on gestational age and weight at birth as well as functional outcomes in the first days of life, however these models are limited to predicting BPD severity in the short term and do not focus on long-term outcomes or risk of developing complications as a result of BPD status (31). Large scale longitudinal studies that would elucidate the long-term impacts of BPD are strongly encouraged in order to validate our findings. We recognize such a study would be a more time and resource intensive than our simulation approach, however based on the magnitude of health and cost burden BPD appears to present, we do believe there is value in a more intensive observational study.
Not surprisingly, our study shows that expected costs, QALYs, and complication rate vary significantly by gestational age at birth or BPD severity. The published literature presented complication prevalence by BPD severity. Our study adds a new contribution to the existing literature by revealing the long-term clinical and economic outcomes by gestational age and BPD severity. A crucial question that remains is whether expected costs and QALYs in fact present as forecasted or if there are latent costs and effects on life expectancy that are expressed later in life, beyond what observational data currently captures. Long-term costs and health utilities for particularly severe complications are likely to be associated with higher risk of death, suggesting some simulants of this model that have severe complications and/or multiple complications have over-estimated costs (due to near end of life care currently unaccounted for) and quality of life measures. It is our hope that further longitudinal research can elucidate the risk-adjusted mortality rate of patients in order to provide better information on disease and care trajectories of BPD. Our simulation model can be extended to test what relative risk increase in mortality would be necessary to see significant changes in our outcome estimates, however this is better explored in the context of an economic evaluation of treatments for BPD.
The best available literature using the same Canadian health system perspective on the long-term healthcare utilization for BPD patients is a 2008 paper by Landry et al. where they estimate the cost of BPD per patient to be $15,700 (adjusted to 2018 dollars) for the first 20 years, excluding index admission. In comparison, our model estimates the annual cost per patient is $16,500 over the same time period, excluding index admission. While closely matched, the two estimates derived costs differently. Landry et al. attributed all hospitalizations and pharmaceutical fees by BPD patients while not explicitly incorporating costs of major complications. In comparison, our study attributed the first two years of hospitalizations to BPD and then based costs on annual costs of major complications, which would include hospitalizations, pharmaceuticals, and reimbursable material costs (16-29).
Evidence on short- to medium-term burden of BPD from non-Canadian sources reinforces our study’s findings about the extreme first two years’ cost of BPD followed by consistent costs incurred as a consequence of major complications (1) and scale of cost difference according to BPD severity. For instance, Álvarez-Fuente et al. found the cost for the most severe BPD cases was roughly 50% higher in the first two years than other BPD patients (1); in comparison our model estimated 44% higher cost of severe BPD cases compared to other patients. While it is difficult to directly compare costs across different health systems and patient groups (Álvarez-Fuente et al. observed all preterm infants rather than extreme preterm), it is encouraging to find differential scales of burden according to severity of BPD. Our study advances the evidence based in projecting long-term health outcomes and cost implications, as well as explicitly incorporating major complications to provide clinical detail on the patient distribution and differential effects on lifetime burden of BPD.
This study focuses on lifetime burden of patients within a Canadian health system perspective, presenting some limitations to generalizability under different health system regimes. The precise costs and outcomes for patients will obviously change under different systems of care, however we believe that the broader implications of health outcomes and the incidence of adverse events will not significantly be different in non-Canadian settings. Our study can therefore provide a starting point for a more general estimation of expected burden of care in other settings.
Due to the limitations regarding data on long-term mortality risk among BPD patients, life expectancy and survival curves were not included as primary outcomes of the model at this stage. While we did incorporate a relative risk to the general population mortality rate based on the best evidence for extreme preterm infants (14), this is non-differential across gestational age at birth or BPD status. Additionally, our model does not yet include risk of mortality associated with major complications, which we would expect to impact survival. Consequentially, our life expectancy estimates are likely to be over-estimated. While this has minimal effect on the total cost estimate since the majority of costs are incurred earlier in life, our health utility estimates are correlated with life expectancy and will be over-estimated concurrent to life expectancy after adjusting for utility discounting.
A limitation of our simulation approach is that the initial population of patients is based on a first-order probability density function approach. While the sampling method provided BPD severity distributions that closely resembled real-world evidence, it did not incorporate other patient characteristics such as birth weight and other perinatal conditions that may be important to accurately predicting adjusted mortality and complication risks. While it is important for these factors to be accounted for in future models, we believed it was important to have an initial model that was based on a smaller number of risk factors—in our case, gestational age at birth and BPD severity—to minimize the number of sources of structural uncertainty in our model. In the absence of clear etiological relationships between correlated risk factors, it is difficult to validate whether a simulated biological pathway is genuine—a risk that increases as more complex interactions across several risk factors are introduced to the model (33). For the purposes of describing the burden of BPD, we believe that gestational age is the primary contributor to differential BPD severity distributions within the extreme preterm population as it is highly correlated to birth weight and other functional outcomes.
Another limitation of this study is that the long-term mortality risk for patients is only based on a long-term longitudinal study of preterm infants, which reports adjusted mortality risk according to extreme preterm birth status (<28 weeks gestational age at birth) but no other risk factors. This is a limitation due to this model being specifically designed to describe differential outcomes among BPD patients, yet mortality outcomes are assumed to be constant across severity strata. We would expect that mortality risk would differ according to BPD severity however there is currently no evidence to establish this. Additionally, better evidence may find that BPD severity is not the predominant factor and that instead other differential risk factors such as early lung function and major complications are better predictors of mortality risk. Our model is capable of incorporating such evidence, however given the limited evidence currently available this remains an under-developed area of the model.
Finally, our model assumes that the risk of complication is independent of other complication status except for BPD severity. The same joint distribution of random effects model from the first stage of our model was applied to estimate the risk of complications after controlling for the risk of mortality. A variance-covariance matrix on the relative risk of complication dependent on other complication status was derived to adjust for compounding risk factors however without sufficient cross-correlation data from the published evidence imputation attempts introduced too much variability into the model to be useful.
Our findings highlight the predicted risks and the long-term health care needs for extreme preterm infants (<28 weeks gestational age at birth) given the current standard of care in Canada. Infants who are discharged are expected to have a reasonably high life expectancy, however the high risk of major complications positively correlated with BPD severity results in severe reductions in expected quality of life. Given the extreme cost burden at the earliest stage of life and lifetime negative impact on quality of life, the most promising interventions would be prevention or mitigation of BPD’s effects that result in the most severe forms of chronic lung disease in extreme preterm infants. Our model and study findings can be used to estimate the maximum scope for therapeutic or health system benefits of a new BPD treatment relative to other existing treatments. The model could also inform research and development decisions and help identify patient and intervention characteristics that will make new treatments for BPD reimbursable.