|
|
||||||
Scott
Grosse, Centers for Disease Control and Prevention Data systems must
be designed with decision makers in mind. It
is not sufficient to collect data. Processes
must be in place to analyze data and report results to decision makers in a
timely fashion. Increasingly,
programs are finding linking databases to be important for improving their
effectiveness and efficiency. Data are used by public health
decision makers in several ways. Surveillance
of variations in prevalence, over time or across states, is one use of data.
Surveillance systems also can be used to identify individuals with
conditions to refer them for services. In
newborn screening, this is known as short-term follow-up.
Data are needed to analyze program operations to improve program
management. In public health this
is known as program evaluation, in NBS as quality assurance.
Surveillance and research data can be used to analyze risk factors in
order to develop or target interventions, to estimate health outcomes, and to
evaluate the effectiveness of interventions.
Finally, research and surveillance data feed into policy analyses of the
costs and benefits of interventions. CDC, along with HRSA, is
working to help states strengthen data systems.
In particular, CDC’s Early Hearing Detection and Intervention or EHDI
program is funding 30 states through cooperative agreements to develop and
operate data systems. This includes
supporting states to link hearing screening with other databases.
Why should programs link data? First,
it can improve effectiveness in ensuring that children are screened and that
those who screen positive receive needed follow-up services.
Second, it can improve efficiency, by facilitating cross-checking and
validation of common data elements and by reducing redundant data collection and
burdens on submitters. For example,
much of the information on blood spot cards is also put by hospitals on birth
certificates. If screening programs
could get this information electronically before specimens are processed, either
from vital statistics or directly from hospitals, programs could reduce errors
and also possibly eliminate some data elements from the card. Data linkages are defined as
using individual identifiable information to link records with different types
of data from multiple sources. Examples
of databases that can be linked to newborn screening include vital statistics,
EHDI, birth defects, children with special health care needs, early intervention
WIC, Medicaid, and immunizations. Data linkages and data
integration are often used interchangeably, but there is a distinction.
With linkages, databases remain separate. Linkages can be performed real time, allowing decision makers
to access different types of data for the same children on demand.
With integration, databases are merged, using a single programming
environment. This allows for the
creation of a comprehensive child record. On
the other hand, real-time data linking can allow the creation of a virtual child
record. In addition to these two pure types, many states are adopting
mixed approaches. This entails
creating subsets of data, with integrated modules – e.g., hearing screening
and blood spot screening, which are then linked to other modules.
Challenges to data linkage or integration include the need for a unique
ID for linking records because of the difficulties with probabilistic matching,
resources for implementation of data systems, access to software and shared data
protocols, and security and privacy concerns. Long-term outcomes of newborn
screening can be studied through controlled trials, in which screening and/or
treatment is randomized, or cohort studies based on observational data.
The only randomized controlled trial of newborn screening with long-term
outcomes is the Wisconsin cystic fibrosis newborn screening study, which was
conducted during 1988-94. Long-term
follow-up has revealed significant benefits in nutritional status resulting from
early identification of cystic fibrosis (Farrell et al., 2001) but no
significant differences in lung colonization or disease (Farrell et al., 1997;
2002). Cohorts can be assembled either
prospectively or retrospectively. In
a prospective cohort study, a group of children are enrolled at birth and
followed over time to monitor utilization of health services and to assess
multiple health and developmental outcomes at specified ages.
To minimize selection bias, cohorts should be population-based.
An example of a prospective long-term outcomes study from a
population-based cohort is the New England Congenital Hypothyroidism
Collaborative (1990) study, in which children detected with hypothyroidism
through screening were evaluated at ages 9-10 years in the late 1980s. CDC
currently has cooperative agreements with Colorado, Iowa, and Oregon/Idaho to
set up long-term tracking systems for children identified through specified
newborn screening tests. A
limitation is that cohort studies of cases identified through screening cannot
assess outcomes in the absence of screening.
The Colorado project is funded through CDC’s EHDI and birth defects
programs. The new project with Iowa and Oregon (paired with Idaho) is
specific to tandem mass spectrometry (MS/MS). Retrospective cohort studies
can begin with a group of children for whom outcome measures of interest are
available, which are then linked backwards in time to data on exposures or
interventions at earlier ages. One source of cases is a clinical registry in
which individuals are enrolled upon receipt of a clinical diagnosis.
An example relevant to newborn screening is the Cystic Fibrosis
Foundation Patient Registry, which includes the age of diagnosis.
Collaborative analyses by CDC epidemiologists of these data have
confirmed the findings of the Wisconsin trial regarding comparable pulmonary
outcomes in early and late-identified children with cystic fibrosis (Wang et
al., 2001; 2002). Other sources of cases for a
retrospective cohort study include educational records or developmental
disability surveillance systems. Cases
can be linked to birth certificates, newborn screening records, and stored dried
blood spots. An example is the
linkage of data from CDC’s Metro Atlanta Developmental Disabilities
Surveillance Program with Georgia’s newborn metabolic screening data, which
identified school age children in metro Atlanta who were born in Georgia and who
had previously been detected with a metabolic disorder or sickle cell disease (CDC,
1999; Ashley-Koch et al., 2001). A retrospective cohort study
can also be based on an inception cohort of children with a common exposure who
are linked forward in time to data on subsequent outcomes.
An example is a CDC-sponsored study of outcomes of sickle cell disease
(SCD) in California, Illinois, and New York in the early 1990s.
Each state participating in the study linked newborn screening records
for all children with SCD with state death certificate data.
The pooled analysis found that children identified with SCD had death
rates between birth and 3 years of age no higher than the general population of
African-American children (CDC, 1998; Olney, 2000). Finally, data from surveillance
and research studies can be used as inputs to quantitative policy analyses that
compare health outcomes, intervention costs and averted costs to determine
whether screening is justified on economic grounds.
Two types of economic evaluation are commonly performed. In cost-effectiveness analysis, health outcomes are either
left in natural units or converted to quality-adjusted life years or QALYs. In
cost-benefit analysis, outcomes are converted to dollars. Economic evaluations can be
performed ex ante, prior to the introduction of an intervention, in which
case model parameters must be based on assumptions informed by the scientific
literature and expert opinion. They can also be performed ex post, after
an intervention has been implemented. In
this case, actual program data on short term outcomes and costs can be used,
although typically assumptions about long-term outcomes with and without the
intervention must be assumed. Three recent economic
evaluations of MS/MS in newborn screening have been presented.
Two were cost-effectiveness analyses, in which cost per QALY ratios were
calculated. One came from Kaiser
Permanente of Northern California (Schoen et al., 2002) and was an ex ante
analysis. The other came from
Wisconsin’s newborn screening program (Insinga et al., 2002) and was an ex
post analysis. A third,
cost-benefit analysis of screening for MCAD alone was prepared for the
Washington state newborn screening program and later adapted for the Arizona
newborn screening program. Two
posters at this symposium discussed this ex
ante model (Thompson et al; Green et al.). The three models differ with
regard to their parameters relative to MCAD.
The estimates of the positive predictive value of MCAD screening are 4%
(Schoen et al.), 25% (Thompson et al.), and 78% (Insinga et al.).
The percentage of children identified with MCAD through screening who are
assumed to die without screening was 16% in the WA model, based on clinical data
from the UK (Pollitt and Leonard, 1998). Insinga
et al. adjusted this percentage to 8% in their base case analysis, based on an
assumption that half of children identified with MCAD by screening would never
have presented clinically. Finally,
Schoen et al. state that they assumed 2.5% mortality in untreated MCAD, although
an even lower rate may have been used for the calculations. In conclusion, better data on both short-term and long-term outcomes are needed to inform newborn screening policy decisions. Rigorous assessment of existing data is also needed, with the assumptions of models subjected to open peer review. Methods of valuation of health outcomes also need to be updated. The QALY weights used in both published cost-effectiveness analyses for MS/MS relied on utility weights for adult victims of neurological disorders, which have not been validated for use with children with developmental disabilities.
|
||||||