Screening Tools
Screening tools were chosen from the CAPHC toolkit for inclusion based on two criteria: i) the cost of administering the screening tool was available or a reasonable approximation could be estimated and ii) an estimate of the diagnostic accuracy, referring to the sensitivity and specificity of the screening tool to FASD, was available or a reasonable approximation could be estimated. Sufficient information on the diagnostic accuracy of the Maternal Drinking Guide Tool, the Medicine Wheel, and the Asante Screening Tool was not identified.
Information was identified to assess the cost-effectiveness of screening tools for two scenarios. The first compares screening newborns suspected of FASD via meconium testing prior to diagnostic testing versus no screening but diagnostic testing for all newborns suspected FASD. Meconium testing screens fecal matter that accumulates over the second and third trimesters for chemical signatures of ethanol that can be indicative of prenatal alcohol exposure [17]. The second compares screening five year olds with the NST prior to diagnostic testing versus no screening but diagnostic testing for all children suspected of FASD. The NST is a questionnaire that asks caregivers about their child’s FASD associated behaviors and risk factors. Depending on the number of positive responses, the questionnaire recommends diagnostic testing or no diagnostic testing [18].
The present study does not directly compare the cost-effectiveness of meconium testing to the NST as these tools are not directly comparable due to the age groups for which they are intended. Meconium testing is for newborn populations and the NST is meant to be used in school age populations.
Cost-effectiveness Analysis
In the present study, incremental cost-effectiveness ratios (ICER) are used to assess value for money [19]. ICER are calculated by dividing the difference in cost by the difference in effectiveness between two interventions. To assess the value for money of an intervention relative to another, decision-makers can compare ICER values to the amount their jurisdiction would be willing to pay (WTP) to gain or willing to accept (WTA) to forgo the effectiveness outcome included in the ICER. Costs are expressed in 2017 Canadian dollars and reflect the perspective of the public healthcare payer. To calculate ICER, a Markov model was constructed. Model parameters such as costs and estimates of the accuracy of screening tools were informed using published literature, expert opinion, and in some cases assumptions based on approaches previously undertaken within the literature.
Model
A hypothetical cohort of children suspected of FASD are evaluated using two versions of an economic model: i) compares screening with meconium testing to a no screening strategy where all children suspected of FASD receive diagnostic testing and ii) compares screening with the NST to a no screening strategy where all children suspected of FASD receive diagnostic testing. Using the model, the total cost and the number of years with an accurate FASD diagnosis were tracked for each strategy until children reached 18 years of age and these values were used to calculate ICER.
Though quality adjusted life years (QALY) are commonly recommended for use as the primary outcome measure in economic evaluations [19], at present the impact of an FASD diagnosis on patients’ health related quality of life (HRQoL) is not understood [20]. As a substitute to QALY, the number of diagnoses or other diagnoses based outcomes are sometimes used as a primary effectiveness measure in CEA of screening strategies [21]. This study uses years with a diagnosis instead of the number of diagnoses, as the former better reflects the temporal nature of a diagnosis. A diagnosis happens at single point in time but has long-term implications. The discounted present value of benefits using the dynamic outcome of years with an accurate FASD diagnosis better accounts for the temporal nature of the decision problem.
In the model, screening can result in true/false positives and patients receive diagnostic testing or true/false negatives and patients do not receive diagnostic testing. Diagnostic testing in both the screening and no screening strategies can result in an FASD diagnosis or no FASD diagnosis. Since newborns who screen positive for prenatal alcohol exposure via meconium testing do not often receive diagnostic testing immediately, the model applied a five-year lag between positive screen with meconium testing and patients receiving diagnostic testing. Figure 1 shows the Markov diagram for the model and Figure 2 shows the decision tree that informed the initial distribution of the hypothetical cohort between states.
In the model, screening can influence costs by reducing the number of individuals who go on to receive diagnostic testing and by affecting the ratio of diagnosed to undiagnosed patients, which are assumed to have different costs and mortality. The study assumes that a year of life with a diagnosis is associated with better outcomes for patients with FASD than a year of life without a diagnosis [1, 22]. A half-cycle correction was applied. The half cycle correction is applied so that patients’ transition through Markov states reflects a mid cycle transition versus at either the beginning or end of the Markov cycle [23].
Model Parameters
Hypothetical Cohort
The present study assumes a hypothetical cohort of 100 children of which 66% (SD=1.4%) meet the criteria for FASD. This represents the percentage of children who received an FASD diagnosis after receiving consultation for suspected FASD at specialized clinics in Western Canada in 2005 [24]. The cohort was assumed to consist of 50% females. This assumption was justified based on the most recent study within Canada on FASD prevalence, which showed similar rates of FASD between males and females [3].
Diagnostic Accuracy of Screening Tools
Meconium Testing
Only studies that included newborns with alcohol exposures at any time during gestation were considered for inclusion, as some identified studies reported newborns with alcohol exposures in the second and third trimesters only. Such studies would inflate the diagnostic accuracy of meconium testing by censoring first trimester alcohol exposures, which can be the most impactful, that meconium testing will not detect. If studies reported multiple criteria for a positive screen on the same group of patients, the strategy with the highest sensitivity was incorporated in the analysis. This is based on CAPHC’s assertion that screening tools should be liberal in their selection for diagnostic testing. From the findings of a recent systematic review [17], four relevant studies were identified [25-28] (Table 1). It should be noted that identified studies reported the diagnostic accuracy of meconium testing to prenatal alcohol exposure and not positive FASD diagnoses. Study results were pooled using a random effects approach [29]. The mean sensitivity was 92.4% (SD=8.1%) and the mean specificity was 51.5% (SD=19.7%).
Table 1. Diagnostic Accuracy of Meconium Testing
Study
|
Study Information
|
Criteria for Positive Screen
|
Sensitivity (SD)A
|
Specificity (SD)A
|
1. Bakhireva et al., 2014
|
Sample Size = 60
Positive Cases Included = 28
Positive Cases = ≥ 0.21 oz alcohol/day at enrollment or ≥ 2.0 oz of alcohol/drinking day.
Controls = No binge drinking in the periconceptional period; ≤ 0.14 oz alcohol/day in periconceptional period; and no drinking at enrollment.
FAEEs Tested = Ethyl Palmitate, Ethyl Stearate, Ethyl Oleate, Ethyl Linoleate.
Limit of Detection = 50ng/g
|
> 600 ng/g all four FAEEs to meconium.
|
100% B (1.9%)
|
13% (5.9%)
|
2. Ostrea et al., 2006
|
Sample Size = 124
Positives Cases Included = 93
Positive Cases = Mothers who used alcohol at the time of conception and/or any time during pregnancy.
Controls= Mothers who reported no alcohol intake around the time of conception or in pregnancy.
FAEEs Tested = Ethyl Myristate
Limit of detection = 50ng/g
|
> 50 ng/g ethyl myristate to meconium.
|
68% C (4.8%)
|
29% C (8.0%)
|
3. Bearer et al., 2003
|
Sample Size = 27
Positives Cases Included = 21
Positive Cases = ≥ 1.0 oz. alcohol/day or ≥ 2 incidents of binge drinking/month in the first trimester of pregnancy.
Controls = Mothers who abstained from drinking during pregnancy.
FAEEs Tested = Ethyl Oleate
Limit of Detection = NA
|
> 13 ng/g ethyl oleate to meconium.
|
100% B (2.1%)
|
67% (17.8%)
|
4. Chan et al., 2003
|
Sample Size = 200
Positive Cases Included = 17
Positive Cases = Mothers who reported any drinking in pregnancy.
Controls = Mothers who reported no drinking in pregnancy.
FAEEs Tested = Ethyl Palmitate, Ethyl Stearate, Ethyl Oleate, Ethyl Linoleate.
Limit of Detection = 50ng/g
|
> 600 ng/g all four FAEEs to meconium. C
|
100% B (2.3%)
|
98% (1.0%)
|
A If SD were not reported, they were calculated using the beta distribution variance formula.
B Sensitivity and specificity were assumed to be 99% instead of 100%, as the beta distribution calculates a variance of 0 for mean values of 100%.
C Estimates were taken from a systematic review and not reported in the corresponding study.
The NST
As with meconium testing, if studies reported multiple criteria for positive screens on the same group of patients, the strategy with the highest sensitivity was incorporated in the analysis. From the findings of a recent systematic review [18], four studies reporting the diagnostic accuracy of the NST were identified [30-33] (Table 2). Study results were pooled using a random effects approach [29]. The mean sensitivity was 85.9% (SD=5.5%) and the mean specificity was 72.9% (SD=10.7%).
Table 2. Diagnostic Accuracy of the Neurobehavioral Screening Tool
Study
|
Study Information
|
Criteria for Positive Screen
|
Sensitivity (SD) A
|
Specificity (SD) A
|
1. LaFrance et al., 2014
|
Sample Size = 80
Positives Included = 48
Positive Cases = Children with FASD diagnosis.
Controls = Typically developing children.
Average Age = 12
|
≥ 6 of items 1-7 or ≥ 3 of items 1-4.
|
63% (6.9%)
|
100% B (1.7%)
|
2. Breiner et al., 2013
|
Sample Size = 60
Positives Included = 17
Positive Cases = Children with FASD diagnosis.
Controls = 18 children suspected for FASD but for whom diagnosis could not be confirmed and 25 typically developing children.
Median Age = 5 C
|
≥ 5 of items 1, 2, 4-8.
|
94% (5.6%)
|
96% (3.0%)
|
3. Nash et al., 2011
|
Sample Size = 109
Positives Included = 56
Positive Cases = Children with FASD diagnosis.
Controls = Typically developing children.
Average Age = 10
Sample Size = 106
Positives Included = 56
Positive Cases = Children with FASD diagnosis.
Controls = Children with ADHD diagnosis.
Average Age = 10
|
≥ 3 of items 1-10
≥ 2 of items 1, 4, 8, 9, 10.
|
98% (1.9%)
89% (4.1%)
|
42% (6.7%)
42% (6.9%)
|
4. Nash et al., 2006
|
Sample Size = 60
Positives Included = 30
Positive Cases = Children with FASD diagnosis.
Controls = Typically developing children.
Median Age = 11 C
Sample Size = 60
Positives Included = 30
Positive Cases = Children with FASD diagnosis.
Controls = Children with ADHD diagnosis.
Median Age = 11 C
|
≥ 6 of items 1-7
≥ 3 of items 1, 4, 8, 9, 10.
|
86% (6.2%)
81% (7.0%)
|
82% (6.9%)
72% (8.1%)
|
A If SD were not reported, they were calculated using the beta distribution variance formula.
B Sensitivity and specificity were assumed to be 99% instead of 100%, as the beta distribution calculates an SD of 0 for mean values of 100%.
C If the average age of study participants was not provided, the median was reported.
Accuracy of Diagnostic Testing
The present study assumes perfect accuracy for diagnostic testing. As it is likely that missed diagnosis of FASD occur, this assumption is assessed in one-way sensitivity analysis. Assuming perfect diagnostic accuracy of diagnostic testing has been undertaken previously in the literature when the sensitivity and/or specificity of a diagnostic test is not known [21]. A benefit of this approach is that it applies a best-case scenario for the accuracy of diagnostic testing in the model. This can help contextualized sensitivity analysis surrounding the parameter. In the present study, this assumption has the effect of biasing results against the screening strategies.
Cost of Screening Tools
Meconium Testing
A cost of $175 was used to approximate the cost of meconium testing [15]. This value was based on the price of meconium testing charged to patients at the Hospital for Sick Children (Toronto, Ontario, Canada) and taken directly from a previous study Hopkins et al., (2008) [15]. .
The Neurobehavioral Screening Tool
To estimate the cost of administering the NST, the present study included time spent interacting with caregivers, time required to administer the NST, and an estimate of the cost of relevant overhead (office supplies, printing services, technology etc.). This included 15 minutes of a social worker’s time, 7.0 minutes of a psychologist’s time, and $5.00 in overhead costs. This corresponded to an estimate of $20 per NST administered. The cost of health providers’ time was based on reimbursement within Ontario, Canada.
Cost of Diagnostic Testing
The cost of diagnostic testing, which includes a physical examination, dysmorphology assessment, neurobehavioral assessment, and prenatal exposure to alcohol confirmation was estimated to be $3,870 [12].
Cost of Health Services Use
First Year of Life
To approximate the annual cost of health service use by patients with FASD diagnoses, this study relies on the work of Stade et al., (2006), who report societal costs for a group of patients of average age 12.9 years with FASD diagnoses [34]. For the first year of life, Stade et al., (2006) report a cost of $20,265 for health services spending. These costs reflect health service utilization related to managing early life medical complications associated with FASD such as low birth weight or prematurity. This cost was applied to the first year of life for the undiagnosed and no FASD groups as well.
Diagnosed FASD
For all subsequent years, a cost of $4,346 per year was applied to the diagnosed FASD population based on the work of Stade et al., (2006). This cost included doctor visits, hospitalizations, emergency department visits, medications, diagnostic tests, and medical devices [34].
Undiagnosed FASD
A lack of information on the cost of healthcare service utilization for undiagnosed patients is often a limitation in CEA of screening strategies [21], as costing studies are not often undertaken in undiagnosed populations. As a result, CEA in screening strategies often need to make assumptions, to approximate costs for undiagnosed populations. To estimate the cost of undiagnosed FASD after the first year of life, this study combines the work of Stade et al., (2006) and McLachlan et al., (2015) [34, 35]. McLachlan et al., (2015) conducted a chart review to investigate the medical, educational, and social services recommended to a group of 70 children assessed for FASD. Of these children: 45 received a diagnosis of FASD; nine had their diagnosis deferred; and FASD was not diagnosed in 16. A deferred diagnosis indicates that FASD could not be confirmed but the diagnostic team was unwilling to rule out FASD. Subsequently, future reassessment is recommended. Though not significant at standard levels (χ2 = 1.48; p-value = 0.223), McLachlan et al., (2015) found that deferred children were 22.2% less likely to be recommended psychiatric treatment than children with a diagnosis. Assuming that service use associated with deferred patients reflects that of patients with undiagnosed FASD and combining this data with the cost reported in Stade et al., (2006) for diagnosed FASD, results in an estimated annual cost of undiagnosed FASD of $3,441 (For further details see Appendix). This assumption is tested in one-way sensitivity analysis. Research suggests that a majority portion of differed patients will go on to receive an FASD diagnosis at some point in their life [22].
No FASD
Patients without FASD were assumed to use healthcare resources at a rate of $3,101/year. This value was calculated using the same method as the cost for undiagnosed FASD [34, 35].
Rate of Future Diagnosis
Patients with FASD who do not receive a diagnosis due to a false negative in screening are assumed to receive future diagnoses at a rate of 5% per year. At present, the rate of future diagnosis for patients with FASD who fail to receive a diagnosis because of a false negative in screening is not known. As a result, this parameter has been estimated based on the assessment of the present study’s authors. This assumption is assessed in one-way sensitivity analysis. These patients are assumed to receive repeated screening and diagnostic testing during subsequent diagnoses.
Mortality
Based on the findings of a recent systematic review [36], one study has reported mortality in FASD [6]. Burd et al., (2008) report a standardized mortality ratio (SMR) for a cohort of individuals diagnosed with FASD of 3.15 [6]. Mortality was assumed to be elevated 10% in the undiagnosed FASD population and to reflect that of diagnosed FASD in the no FASD population. At present, the rate of mortality for undiagnosed patients is not known and this parameter was informed based on the assessment of the present study’s authors. Mortality assumptions were assessed in one-way sensitivity analysis. SMR were combined with Statistics Canada life tables to estimate mortality.
Discounting and Time Horizon
Cost and outcomes occurring beyond one year were discounted at a rate of 1.5% [19]. Discounting weights events occurring sooner to a greater extent than those occurring later to account for societal preference for the present. Costs and outcomes were aggregated until children reached age 18. This time-horizon was chosen as consultation with experts suggested that pediatric diagnosis is of greater value for improving patient outcomes than diagnosis in adulthood. Additionally, as there are few treatment options available for adults [9], it is not clear how service utilization between diagnosed and undiagnosed adults would differ.
Probabilistic Analysis
To conduct probabilistic analysis (PA), values were randomly sampled for each model parameter from a distribution and then used to calculate ICER. This process was repeated 5,000 times using Microsoft Excel (Microsoft Corporation, Redmond, WA, USA).
For parameter values that represent percentages, a beta distribution was applied with mean and SD based on literature-derived estimates. Two exceptions to this are i) the rate of subsequent diagnoses was varied subject to a uniform distribution over the range 3% to 7% and ii) the mortality for undiagnosed FASD was varied subject to a uniform distribution by an increase of 0 to 20% relative to the mortality of diagnosed FASD. At present uncertainty for the aforementioned parameters (i and ii) is not well understood, the authors of the present study chose these intervals to reflect a large degree of uncertainty for these parameters. The uniform distribution was chosen, as it makes each value within the PA interval equally likely further accounting for uncertainty. Cost for screening tools and diagnostic testing were varied subject to the normal distribution within the interval plus or minus 25% of the parameter value with an SD of 10% of the parameter value. Estimates for costs based on experimental results were varied subject to the log-normal distribution with an SD of 10% of the parameter value. Parameter values for mortality were varied subject to the normal distribution but it was assumed that mortality could not be superior to the general public. The SD for the mortality of patients with no FASD represents the SD of diagnosed FASD inflated by 1.25, to account for uncertainty regarding mortality rates in this population. Sensitivity and specificity estimates were not correlated in PA. This may result in the model overstating uncertainty. For a list of model parameters and distributional assumptions, see Table 3.
Table 3. Parameter Values, Standard Deviations, and Distributional Assumption
Parameter and Reference
|
Mean (SD)
|
Distributional Assumption
|
Hypothetical cohort characteristics
|
|
|
- % Positive cases24
- % Female
- Age screened Meconium Testing
- Age screened NST
|
66.3% (1.4%)
50.0%
Birth
5 years
|
Beta
Not varied
Not varied
Not varied
|
Diagnostic accuracy of screening tools
|
|
|
- Meconium testing
Sensitivity25-28
Specificity25-28
- The NST
Sensitivity30-33
Specificity30-33
|
92.4% (8.1%)
51.5% (19.7%)
85.9% (5.5%)
72.9% (10.7%)
|
Beta
Beta
Beta
Beta
|
Accuracy of Diagnostic Testing
|
|
|
- Sensitivity
- Specificity
|
100%
100%
|
Not varied
Not varied
|
Cost of screening tools and Diagnostic Testing
|
- Meconium testing15
- The NST LD
|
$175 ($18)
$20 ($2)
|
Normal (bounded ± 25% of mean)
Normal (bounded ± 25% of mean)
|
- Cost of diagnostic testing12
|
$3,870 ($387)
|
Normal (bounded ± 25% of mean)
|
Annual Cost of Healthcare Service Use
|
|
|
- First year of life34
- Diagnosed FASD34
- Undiagnosed FASD34,35
Diagnosed recommended to receive psychiatric care35
Undiagnosed recommended to receive psychiatric care35
- No FASD34,35
|
$15,976 ($1,598)
$3,426 ($343)
$2,713
55.6% (7.3%)
33.0% (14.7%)
$3,101
|
Log-normal
Log-normal
Varied based on inputs A
Beta
Beta
Not varied
|
Future Diagnosis Rate
|
|
|
- Rate of future diagnosis for undiagnosed patients AB
|
5%
|
Uniform (bounded ± 2%)
|
Mortality
|
|
|
- Diagnosed FASD6
- Increased mortality for undiagnosed FASD relative to diagnosed AB
- No FASDAB
|
3.15 (1.6)
10%
3.15 (2.0)
|
Normal
Uniform (bounded ± 10%)
Normal
|
The values for the Annual Cost of Healthcare Service Use in Table 3 reflect that prior to adjusting for inflation. The Annual Cost of Healthcare Service Use parameters were varied prior to adjusting for inflation and then inflated for probabilistic analysis.
A Inputs refer to Diagnosed recommended to receive psychiatric care and Undiagnosed recommended to receive psychiatric care.
LD Parameter was informed with unpublished local data.
AB Parameter was informed based on authors’ assumption.
One-way Sensitivity Analysis
To conduct one-way sensitivity analysis, parameter values for key model inputs were varied by plus and minus 25% of the parameter value. Parameters included in one-way analysis include the sensitivity and specificity of screening tools, the sensitivity of diagnostic testing, the number of positive cases in the cohort, the annual cost of diagnosed and undiagnosed FASD, the cost of diagnostic testing, the cost of screening tools, future diagnosis rates, and mortality rates. Alternative discount rates were assessed in scenario analysis based on the recommendation of CADTH [19].