Comprehensive Cost-Utility Analysis of Newborn Screening Strategies

Study Design

We developed a decision model with decision analysis software (DATA 4.0; TreeAge Software, Williamstown, MA), to compare various strategies for screening for newborn diseases. We adopted the recommendations of the Panel on Cost-Effectiveness in Health and Medicine for conducting and reporting a base-case analysis.¹⁷ We assumed a societal perspective to produce results, permitting comparisons across different interventions, and expressed results in terms of discounted costs, quality-adjusted life-years (QALYs), and incremental cost-effectiveness ratios.

Decision Model Structure and Assumptions

To frame the model for analysis, we made a number of simplifying assumptions. We assumed that any child could have, at most, 1 of the conditions for which the tests screen. We used a multiplicative utility model, that is, for outcomes with >1 complication, we multiplied the quality adjustments (utilities) for all relevant complications together. We assumed the existence of a comprehensive screening program that ensures timely testing, result notification, and appropriate treatment of all positive screens. Finally, we assumed that MS/MS methods would be used to screen for multiple conditions.

The Decision Model

Figure 1 shows the decision model structure. The decision tree consisted of a decision node whose branches represent 8 screening tests, that is, PKU, CAH, CH, biotinidase deficiency, MSUD, galactosemia, homocystinuria, and MS/MS tests. We assumed that MS/MS would be used to detect PKU, biotinidase deficiency, MSUD, galactosemia, and homocystinuria, as well as MCAD deficiency. In a second analysis, we compared MS/MS directly with a screening panel that included PKU, biotinidase deficiency, MSUD, galactosemia, and homocystinuria detection with conventional methods.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1

Decision model. Branches from the square decision node represent the various screening tests. Following each of these is a subtree representing each of the conditions that might be present. Only one of these subtrees is shown. Following each condition is a subtree depicting the possible outcomes of the conditions. Final outcomes are shown for homocystinuria (HCY) only. # indicates the probability of the lower branch is 1 minus the probabilities of the branches above. BIOT indicates biotinidase deficiency; GAL, galactosemia; Dz, disease.

Following each test branch is a chance node whose branches represent the 8 conditions that the child may have, ie, PKU, CAH, CH, biotinidase deficiency, MSUD, galactosemia, homocystinuria, or MCAD deficiency, or “no disease.” Each disease branch is followed by a subtree that represents potential sequelae of the disease. In Fig 1, final outcomes are shown only for homocystinuria. PKU was associated with a risk of mild, moderate, or severe developmental delay and/or seizure disorder. CAH could lead to early childhood death (average life expectancy: 1 year). CH could cause mild, moderate, or severe delay. Biotinidase deficiency could lead to any combination of seizure disorder, severe developmental delay, hearing loss, or blindness. MSUD could lead to any combination of severe or mild cerebral palsy (CP) or mild, moderate, or severe developmental delay. Galactosemia could lead to death in infancy or moderate or severe delay. Homocystinuria had a risk of mild, moderate, or severe delay and/or vision loss. MCAD deficiency could lead to early death (modeled as death at 1 year, on average) or CP and/or mild-to-severe developmental delay. Children with no disease face a risk of a false-positive newborn screening result and its attendant expenses.

The probability of each of the outcomes with screening, P(outcome|screen), was represented by the formula P(outcome|screen) =P(outcome) × [1 − (efficacy × sensitivity)], where P(outcome) is the probability of the outcome (eg, CP or delay), without screening or treatment, derived from the literature (see below), efficacy is the efficacy of early detection and treatment in preventing the outcome, and sensitivity is the sensitivity of the screening test for the disease.

Data and Assumptions

Probabilities of sequelae, quality-adjusted survival rates, and costs for treatment were derived from various clinical and administrative data sources. We searched Medline (from January 1980 to November 2002) for studies with the terms “newborn screening” and any of the disease entities we were investigating. We also searched online Internet databases for the same terms. We supplemented this strategy by hand-searching citation lists from articles identified in the electronic search and recent reviews and cost-effectiveness analyses of other newborn screening strategies.

The prevalence of each disease was estimated, when possible, from the National Newborn Screening Report.¹⁸ Supplemental information was taken from the American Academy of Pediatrics newborn screening fact sheets and secondary analyses.^8,19 We determined the sensitivities and specificities of the various screening tests through national reports and other secondary analyses.^10,13,18,20 Probabilities of various sequelae and the effectiveness of early intervention in preventing these sequelae were also estimated from the newborn screening fact sheets, secondary data analyses, and cohort studies of patients with the diseases in question.^19,21–28 Base-case probabilities for major complications are summarized in Table 1.

Estimation of Life Expectancy

We estimated the average life expectancy of normal patients from the National Vital Statistics Reports published by the Centers for Disease Control and Prevention.²⁹ We estimated the effect of disability on life expectancy through a cohort study on the subject.³⁰

Quality-of-Life Adjustments

We adjusted life expectancy for quality of life by using health state utilities.¹⁷ We calculated QALYs by multiplying remaining years of life by the utility value for each health state. Utilities for most disabilities were drawn from a study assessing parents' utilities by using a standard gamble, adjusting for different severities when necessary.³¹ The study assessed utilities for outcomes of occult bacteremia and meningitis, which resemble closely outcomes of the diseases for which newborns are screened. Specifically, parent utilities were assessed for deafness and various degrees of developmental delay and CP. Respondents in the study were parents of children between 2 and 24 months of age who visited an urban emergency department. These were the most relevant utilities available for these outcomes. Utilities for blindness were taken from a study that developed an equation for estimating utility on the basis of visual acuity.³² We did not adjust utility if a disease state did not result in a disability that was thought to affect quality of life. Base-case estimates for utilities are summarized in Table 1.

Costs

The costs for the specific screening tests were derived from 2 sources, ie, a PriceWaterhouseCoopers analysis of newborn screening costs and a prior cost-effectiveness study of MS/MS screening.^9,33 These costs incorporated amortized start-up costs and operating costs. The costs of treating selected diseases over the course of a lifetime were estimated with data from the Office of Technology Assessment analysis of strategies for newborn screening, data from a Washington State report on the cost-effectiveness of some screening strategies, and, in 1 case, expert opinion.^21,34 Costs of treating sequelae of different diseases were estimated from sources including a Morbidity and Mortality Weekly Report on economic costs associated with mental retardation, CP, hearing loss, and visual impairment; an epidemiologic study of end-of-life care costs; and prior secondary analyses.^14,21,35,36 All costs were adjusted to 2004 US dollars with an inflation calculator on the Web site of the Department of Labor, Bureau of Labor Statistics.³⁷ For this study, costs were a complex mixture of medical and nonmedical costs, and we elected to adjust costs with the General Consumer Price Index, rather than the Medical Care Component Consumer Price Index. Base-case estimates for costs are summarized in Table 1.

Time Preference

To reflect patient preferences for health outcomes and having material goods sooner rather than later, we discounted all costs and health effects at an annual rate of 3%.

Calculation of Incremental Cost-Effectiveness Ratios

For each treatment strategy, we calculated the expected total costs by multiplying the probability of each unique outcome with its associated costs and then adding these values for all possible outcomes. Total QALYs for each treatment were calculated in a similar manner. We then calculated the incremental cost-effectiveness of each screening strategy by dividing the difference in costs by the difference in QALYs.

Sensitivity Analysis

We used 1-way sensitivity analysis to identify important model uncertainties. All probabilities were tested across the full range from 0 to 1. Costs were analyzed from $0 to $1 million. To test the impact of changing multiple assumptions, we developed a “pessimistic” case analysis that biased the model against newborn screening. In the pessimistic case, we eliminated indirect costs from the estimated costs of developmental delay and CP.³⁵ We eliminated the costs of developmental delay from any outcomes that included CP. We increased the cost of MS/MS to $20, the high end of published estimates. We increased the cost of evaluating false-positive screening test results to $1000. We used low-end estimates for the prevalence of MCAD deficiency (1 case per 18000 infants) and PKU (1 case per 20000 infants). We also used a low-end estimate of the risk of death resulting from CAH (3%).

Base-Case Analysis

The results of the base-case analysis are shown in Table 2. The first column of Table 2 shows the testing strategies in ascending order of average cost. The second column shows the average cost of each strategy. This includes the cost of the test and the average costs (per person screened) associated with all of the treatments and outcomes for the diseases represented in the model. The third column shows the difference in average costs between the testing strategy and not testing. A negative number means the strategy saves money, on average, over not screening. The fourth column shows the average effectiveness of each strategy in QALYs. The fifth column shows the difference in effectiveness between each strategy and not testing. A positive number means there is a gain in QALYs. The sixth column is the average cost-effectiveness ratio for each strategy, and the seventh column is the incremental cost-effectiveness ratio, the incremental cost per QALY gained. Tests marked as “Dominates” save money and improve outcomes relative to not testing.

All except 2 screening tests dominated the no-testing strategy. That is, these screening tests improved outcomes and reduced costs relative to not screening. The 2 exceptions are screening for CAH, which costs slightly more than $20000 per QALY gained, and screening for galactosemia, which costs approximately $94000 per QALY. The screening test with the lowest expected cost is MS/MS.

In addition to comparing each test with a no-testing strategy, we compared a panel of conventional tests for PKU, biotinidase deficiency, MSUD, galactosemia, and homocystinuria with MS/MS methods to screen for the same conditions and MCAD deficiency. The cost and effectiveness comparison is shown in Table 3. When used in this multiplex manner, MS/MS was less expensive and more effective than the comparable panel of conventional screening tests.

Sensitivity Analysis

Because most screening tests were dominant over not screening, we conducted sensitivity analysis on all of the variables in the model, to determine at what levels the tests were no longer cost saving. The fourth column of Table 1 shows the threshold value for each variable, at which the corresponding screening strategy becomes more costly than not screening.

In the base-case analysis, we assumed that the sensitivity of all of the screening tests was 100%. When we varied test sensitivity from 0% to 100%, the dominant strategies remained dominant to sensitivities well below those in the literature, as follows: PKU, 10%; CH, 67%; biotinidase deficiency, 13%; MSUD, 83%; homocystinuria, 37%; MS/MS, 27% (Table 4).

Because of the costs of evaluating false-positive results, the dominant testing strategies became more costly as specificity decreased. When the specificity of a screening test was lowered, a larger proportion of unaffected children had false-positive test results, which resulted in increased costs to rule out disease. When the specificity of a test was below a threshold value, some tests were no longer cost saving (Table 1); however, threshold values were below those reported in the literature.

The prevalence of a disease also affected its incremental cost. Biotinidase deficiency, CH, homocystinuria, MSUD, and PKU all became costly when the prevalence was decreased to values between 3 and 10 times lower than baseline values. When the prevalence of CAH was very high (3.5%), screening for CAH became cost saving. The cost of screening for galactosemia was not sensitive to prevalence, and the cost of MS/MS screening was not sensitive to the prevalence of MCAD deficiency.

The results were relatively insensitive to the rates at which individual sequelae occurred. Screening for CH was not cost saving if the rate of mild developmental delay without treatment was less than one tenth of baseline values or if the risk of severe delay was decreased by approximately one half. If the risk of blindness resulting from biotinidase deficiency was <40%, then screening for biotinidase deficiency alone was no longer cost saving. Screening for MSUD alone was not cost saving if, without treatment, the risk of severe CP was <14% or that of severe developmental delay was <34%. Because MS/MS screens for many conditions, it remained cost saving despite changes in any one of these parameters. Testing for galactosemia had a positive net cost in all sensitivity analyses; however, when the risk of neonatal death was >27%, the cost of screening fell below $50000 per QALY saved.

The cost savings with the screening tests depended in some cases on the effectiveness of treatment in preventing specific sequelae. Screening for CH was not cost saving if it was <67% effective in preventing severe delay. Screening for homocystinuria was not cost saving if treatment was <31% effective in preventing blindness. MSUD screening was not cost saving if treatment was <59% effective in preventing CP or <36% effective in preventing delay.

Our results were not sensitive to the costs of screening tests. Screening test costs that exceeded savings from disease costs were between 2 and 10 times reported costs of screening tests. Conversely, screening for galactosemia or CAH was not cost saving even if tests could be performed for free.

Lifetime costs of diseases and their sequelae are difficult to estimate. However, we found that the cost saving for most of these screening tests was either insensitive or sensitive at values significantly below reported rates.

Our pessimistic case analysis biased the model against screening. Under this set of assumptions, newborn screening with MS/MS methods was not cost saving relative to a conventional screening panel, and the screening panel was not cost saving relative to no screening (Table 5). The conventional screening panel cost $310.56 per QALY saved over not screening. Compared with this panel, MS/MS screening cost $4838.71 per QALY saved.