Objective. To determine coverage of the newborn screening program (NSP) for metabolic disease in Alberta, Canada, and to determine reasons for not being screened.
Study Design. Coverage was estimated by deterministic matching of live birth registration data with newborn screening data for the year 1992. Demographic characteristics of not-matched infants were compared with good-match infants using logistic regression.
Results. For 42 392 live births, there were 41 553 screening records, of which 40 593 infants were very good matches to NSP records. Another 960 were possible matches. A total of 839 infants were not matched at all, and coverage was estimated at 98.0%. Determinants of infant not-matched status were death in week 1 (adjusted odds ratio [OR]: 383); birth weight of <1500 g (adjusted OR: 18.9) or between 1500 and 2500 g (adjusted OR: 3.2); having a mother who was single (adjusted OR: 2.7) or formerly married (adjusted OR: 12.9); or being born out of hospital (OR: 19.2). The calculated 98% coverage is close to an estimate of 98.3% made by the NSP comparing total births with initial screenings.
Conclusion. The matched data give insight as to who was missed and point to the need for closer attention for infants at greater risk of not being screened for metabolic disease.
Since the introduction of the Guthrie test,1 screening of newborns has become an effective method to detect certain metabolic diseases. Disorders commonly screened for include congenital hypothyroidism and phenylketonuria (PKU), but some programs include other disorders.2 These diseases are rare, present at birth, hard to detect clinically in the newborn, and can impair the health and mental development of the child permanently if not diagnosed and treated promptly. Where newborn screening programs (NSPs) exist, it is commonly recommended that all newborns undergo screening at between 1 and 7 days of age.3-5
A primary concern in NSPs is the completeness of coverage; ie, are all newborns screened? This usually is estimated by comparing the number of unique screen samples analyzed with the number of births registered for the same time period and region. Rarely is a formal link made between the two, and even more rarely is any attempt made to determine which children are not screened.
In Alberta, Canada, all newborns, ∼42 000 per year, are screened for PKU, congenital hypothyroidism, biotinidase deficiency, and tyrosinemia. The program is part of the Alberta Hereditary Diseases Program and is carried out by the Department of Laboratory Medicine and Pathology of the Walter Mackenzie Health Sciences Centre in Edmonton, Alberta. Although coverage is estimated to be close to 100%, it has never been evaluated formally.
This project had three objectives: 1) to determine the number of infants born alive in Alberta in 1992 who had blood obtained for newborn screening; 2) to define the characteristics of those infants who were not screened or who could not be identified with confidence as being screened; and 3) to determine the degree to which screened infants with results requiring follow-up were followed up appropriately and a final disposition made. This article addresses the first two objectives.
SUBJECTS AND METHODS
Data came from two sources: 1) the Government of Alberta Register of live births, and 2) the Newborn Screening Program of the Alberta Hereditary Diseases Program. Ethics approval for the project was obtained from the ethics committee of the University of Alberta Faculty of Medicine.
Coverage was estimated by deterministic matching of all registered live births in Alberta in 1992 with screening program data to determine the number of infants screened. Infants remaining not matched (NM) were considered to be unscreened. The demographic and biological characteristics of the unscreened infants were compared with screened infants. Variables used in the analysis were those available from the two datasets plus variables derived from the datasets. Table1 shows the variables available from each of the datasets.
Data preparation consisted of 1) abstracting, cleaning, and verifying the screening data from laboratory files for the period from January 1, 1992, to January 31, 1993 into a working screen file; 2) obtaining from Alberta Registries the birth data file for all live births in Alberta in 1992 and a file for all infants born in 1992 who died in the first year of life and preparing these data for merging with the screen file; 3) estimating a birth date for the screen record based on the date of the report and infant age at that time; and 4) merging the screen file and birth file to make a file suitable for analysis (Fig1).
Data were stored in Paradox6 and Reflex7databases and were analyzed using SPSS.8 Matching was performed with an iterative series of SPSS MATCH procedures and also by visual matching of screen and birth records. Except as noted below, matching was deterministic and based on the last name of infant and mother, sex, date, hospital and community of birth, and normal residence of the infant. In general, for matching purposes the place where blood was obtained was considered to be the place of birth, and the estimated date of birth on the screen record was compared with the registered date of birth. Because a child may take the mother's last (maiden) name or it may have a hyphenated name, visual matching was attempted as the last stage of the matching procedure.
Each matched record was assigned a level of confidence reflecting one author's (D.W.S.) subjective estimate of the quality of the match for that record. An infant had very good match (GM) status if confidence in the match was high and had a possible match (PM) status if confidence was low. The PM infants were chosen using the criteria that the birth record child had to have a birth date closest to but before the estimated birth date of an existing unmatched screen result; had to be alive on the date the blood was recorded as obtained; and had to be from the same community, the next nearest community, or the most likely referral community as the existing unmatched screen result. Cases in which the criteria were equally met by more than one match were assigned a match at random. The children remaining in the birth file after all the screen records had been matched were considered NM; thus there were three categories of infants: GM, PM, and NM.
On completion of matching, unique personal identifiers were deleted and the age when blood was obtained was calculated, using date of birth from the birth certificate, and date of sample from the screen record. The variable native was created by considering that infants with homes on native reserves were very likely of native origin.
The GM and PM groups were combined to obtain an estimate of coverage only. For much of the remaining analysis, the PM group was not included. Descriptive summary statistics were obtained for all variables. The statistical significance of analysis of categorical data was assessed using contingency table analysis and χ2statistics, and of continuous data using analysis of variance with post hoc comparisons using Scheffé's procedure with a probability criterion of P < .05.
The variables birth weight and maternal age also were divided into categories for use in contingency table analysis. Birth weight was divided into <1.5 kg, 1.5 to 2.5 kg, and >2.5 kg categories, and maternal age was divided into <20 years, 20 to 35 years, and >35 years categories.
The choice of variables to explore as determinants of nonmatching was based on availability of data, clinical intuition, and the results of Streetly.9 After initial exploration comparing single variables with match status, logistic regression was used to account for several factors simultaneously, using matching status (GM vs NM) as the dependent variable. Those variables significantly associated with match status in the exploratory analysis were included as independent variables.
Data were available for 43 155 screen records: 41 553 were initial screen results, and 1602 were repeat screen results from 1485 infants. Of the 42 392 live births registered, 40 593 were GM infants, 960 were PM, and 839 were NM. The sum of GM and PM infants (41 553) is the best estimate of coverage and is 98.0% of the eligible population. Almost 75% of all infants were born to married women, 22% were born to women who had never married, and the remaining 3% were born to women who were divorced, widowed, separated, or whose marital status was unknown. For analytic purposes, the mothers in these last four groups were combined to form one category called formerly married. For 1554 infants (3.7%), the probability of native status was high. Of the 167 infants who died in the first week of life, 13 were GM, 14 were PM, and 140 were NM infants.
Two communities, each with >13 000 births per year, accounted for 63% of all births. Communities with between 100 and 1000 births per year accounted for ∼24% of births, and smaller communities accounted for the remainder. There were 226 infants who were born out of hospital (OOH); these were considered home births. A large proportion of infants were born in communities different from their home community.
Data describing the GM, PM, and NM groups are shown in Table2. Compared with the GM group, NM infants more often died in infancy, were of low birth weight, and were born after a shorter gestation.
Nearly 5.4% of NM infants were born OOH, compared with only 0.4% of GM infants. Only 167 (73.9%) of the 226 infants born OOH were matched. As well, NM infants were slightly more likely than were GM infants to have come from a community with a high native population and were more likely to live in a community different from the community or administrative health region in which they were born (data not shown).
Overall, 6.2% of infants had a birth weight <2500 g, and 1.2% had a birth weight <1500 g. However, 32.1% in the NM group and 14.9% in the PM group were <2500 g, but only 5.5% of the GM group was <2500 g. For infants with a birth weight <1500 g, the differences between the NM and GM groups were even more striking (18.6% vs 0.7%). Because so many NM infants died in the perinatal period, birth weights excluding deaths in the first week of life were compared. With first-week deaths excluded, infants <2500 g constituted 24.2% of NM infants compared with 5.5% for the GM group.
Only 50.8% of NM infants had married mothers, compared with 76.0% of GM infants. The infants of formerly married mothers were much more likely to be NM than were those whose mothers marital status was recorded as single. Of the 839 women in the NM group, 14.5% were younger than 21 years, compared with 9.1% of the GM group. Older mothers were generally evenly distributed between the two groups (9.4% and 8%, respectively).
Table 3 summarizes data describing the relationship of match status to maternal age, birth weight, and marital status. Data for infants whose birth weight was missing are excluded. Except for the formerly married, younger than 21-year group, of whom there were only 20 cases, matching was most likely to occur among infants with a birth weight >2500 g born to married mothers 21 to 35 years of age. At every age, as birth weight rises, the proportion of infants matched also rises. The proportion of GM infants rises also as marital status changes from formerly married to single to married. These data suggest that match status was influenced by maternal age, marital status, and birth weight.
Initial exploration of the data using logistic regression eliminated the variables describing probability of native status, maternal age, and mobility as important determinants of match status. Variables included in the final regression were death in week 1, birth weight (three categories), marital status (three categories), and birth OOH. The results of this regression are generally consistent with the stratified analysis performed previously (Table4) but demonstrate the independent effects of the significant variables.
In summary, 98% of infants were screened; however, 839 infants were not screened. Those at greater risk of not being screened were those born OOH, of low birth weight, to unmarried or formerly married mothers, or dying in the perinatal period.
The credibility of this study rests on the assumption that all initial screen records were obtained; that the match was complete and accurate; and that all repeat samples were detected, recorded as such, and eliminated from estimates of coverage. All the screening data were on archive data tapes of the clinical laboratories of the Walter Mackenzie Health Sciences Centre of the University of Alberta. Data abstraction was exhaustive and involved at least four separate scans of the screening data files. The quality and accuracy of the birth data were assumed to be very high, because they were obtained from vital statistics data provided by the Government of Alberta.
Because both the computer and visual match processes were based on specific identifiers such as last name and date and place of birth, it is likely that the accuracy of the match is high. With the assignment to each record of a level of confidence of a match, the subsequent analysis, which is based primarily on records with strong confidence of a very good match, is conservative in its conclusions.
A sample was considered a repeat sample if it was labeled as such or if records with equivalent identifiers were found during data preparation or matching. The method of matching for repeat analyses was almost always a visual match. The search for repeat analyses ended after finding the first repeat, although on occasion several sequential analyses were found for an infant.
Guidelines have been developed by various agencies that describe the institution and management of an NSP.3-5 Although these guidelines differ in specifics, they all emphasize the concepts of 1) centralization; 2) rigorous coverage of all infants at an age >24 hours and <7 days; 3) the use of accepted screening tests for the respective metabolic diseases; 4) a clear line of responsibility for transmission of all screen results to the physician or hospital identified as the sender; 5) a more direct and immediate method of informing physicians of abnormal results; 6) follow-up of suspect results until a diagnosis is made and definitive therapy is instituted; 7) some form of ongoing evaluation, usually participation in national or international quality control programs that allow an assessment of sample processing; and, 8) keeping accurate records.
Estimates of coverage vary among countries and over time, but in established programs coverage often approaches 100%.10-13In the United States, coverage for 1992 varied from ∼76% for Nevada to 100% for several other states. However, Puerto Rico, a US trust territory, screened only 8% of its newborn population.2
Although acceptable for administrative purposes, most estimates of coverage are crude, commonly comparing only total unique screens and total births for the same area and time. With this method, it was estimated that in 1992, 98.3% of infants in Alberta were screened. This study extended the usual estimate of coverage by linking screen records with birth data and determining actual coverage and showed that 98.0% of children were screened at least once. This is in good agreement. However, the relatively low value of 2% nonscreened infants represents 839 infants. It is possible that some of these infants were born in Alberta but moved out of province before being screened; however, this situation was determined in only 27 infants. This does not include an additional 531 infants who were first screened only after 7 days of age, nor does it include 47 infants who had inadequate or unsatisfactory samples but who did not undergo repeat screening (data not shown). The figure of 98% is less than that of 99.1% reported recently by Gray and colleagues,14 who matched birth and screen records for 8751 consecutive births at two large hospitals for part of 1993. However, the data reported here reflect all infants born in one province for 1 year.
Total coverage remains a goal of all NSPs. In Alberta, with 43 000 births year and screen coverage of 98%, on average about once every 5 years, a child with congenital hypothyroidism (incidence, 1:4000)15 will be missed because of a failure to screen. Apart from the personal tragedy that these missed cases represent, they also constitute a significant legal concern. In the United States, up to 1986, of 76 missed cases, 29% resulted in legal action, and settlements ranged from $1 million to $20 million.16
Who Got Missed
A strength of this report is the description of the characteristics of infants who were not screened. The results obtained agree in part with those of Streetly et al,9 who suggested that infants in special care units, infants born in one area but normally lived in another, and infants of high-risk ethnic groups also were more likely to be missed. Gray14 also reported that low birth weight infants were at particular risk of being missed. Although the current data verify that low birth weight infants are at significant risk of being missed, they extend these findings to show that risk of being missed is also associated with single parenthood, home birth, and death in the first week of life. Our results differ from Streetly in that neither mobility nor ethnic (native) status was independently associated with being missed. This may be in part attributable to our inability to identify accurately ethnic status or mobility.
A current concern regarding nonscreening is the practice of discharge from hospital before 24 hours of age. Gray14 showed that infants discharged before 24 hours were 25 times more likely not to be screened than those discharged after 24 hours. We cannot determine with accuracy which children were discharged before 24 hours and who were not screened, thus, we cannot comment about this specific, and important, concern.
Death in week 1 of life is the strongest predictor of nonscreening, with 140 of 167 deaths by age 7 days in the NM group. Because most deaths occurred in the first day of life, this finding is not surprising, although it could be argued that even very sick infants should be screened. The logistic regression showed that independent of death in week 1, very low birth weight infants (<1500 g) and infants with birth weight between 1500 and 2500 g were 19 times and 3 times, respectively, less likely to be screened compared with infants with birth weights >2500 g. No specific reason for this observation is apparent; however, the problem of small preterm babies being missed requires resolution.
Although fewer in absolute number, infants born OOH also were at great risk of not being screened. Of the 226 infants born OOH, 58 were not screened, resulting in a nonscreen rate 19 times that for children born in hospital. Many of those screened were >7 days of age (data not shown), potentially prolonging the time to diagnosis and treatment.
Finally, infants of formerly married mothers or of mothers whose marital status was unknown also were at high risk of not being screened. Infants of single mothers were at lesser risk, and infants of mothers who were married were at least risk.
An unexpected result is the marked difference in the odds ratios (ORs) for being NM for infants of mothers with formerly married marital status (OR: 12.9) compared with infants of single mothers (OR: 2.7). It seems reasonable that infants of mothers in these two groups might have a greater frequency of being NM than do infants of married mothers, but it is not obvious why infants of formerly married mothers are at such increased risk. A possible reason is confusion in surnames of the infants as recorded when screened and on the birth registration form. Although visual matching was performed as part of the matching procedure and every effort was made to detect alternate names, it is possible that some were missed.
Some children with disease will be missed in screening programs. This may be attributable to the vagaries of the disease or because the screening tests are not perfect.17 However, the primary determinants of an affected child not being detected are that the child was never screened; the blood sample obtained was misinterpreted, unsatisfactory, or obtained too early in the child's life; or follow-up for repeat blood analysis was incomplete.18-20 This report has addressed the first reason. Of the 781 infants not matched and for whom all the data in Table 4 were available, 609 had one or more risk factors as determined by logistic regression. Among 11 772 infants with one risk factor, 434 (3.7%) were missed, and among 971 infants with two or more risk factors, 175 (18%) were missed. However, among 29 585 infants with no risk factor, only 172 (0.58%) were missed.
This work was supported by the Department of Pediatrics of the Faculty of Medicine of the University of Alberta.
We thank Kathy Mazurek of the Department of Clinical Laboratories and Mr Harry Calhoun of the EPICORE project for facilitating the access to data, and David Spady who wrote the necessary computer programs to extract the screening data from the laboratory data archive.
- Received November 20, 1997.
- Accepted April 8, 1998.
Reprint requests to (D.W.S.) Department of Pediatrics, 2C3.62 WMC, University of Alberta, Edmonton, Alberta, Canada T6G 2R7.
This research was carried out as a project in partial completion of the requirements for an MSc (Public Health Sciences) for Dr Spady.
- PKU =
- phenylketonuria •
- NSP =
- newborn screening program •
- NM =
- not matched •
- GM =
- good match •
- PM =
- possible match •
- OR =
- odds ratio •
- OOH =
- out of hospital
- Guthrie R,
- Susi A
- ↵Newborn Screening Committee. The Council of Regional Networks for Genetic Services (CORN) National Newborn Screening Report—1992. Atlanta, GA: CORN: 1995
- American Academy of Pediatrics, Committee on Genetics
- LaFranchi S,
- Dussault JH,
- Fisher DA,
- et al.
- ↵Paradox. Borland Paradox for Windows. Version 5.0. Scotts Valley, CA: Borland International, Inc; 1994
- ↵Reflex: Borland Reflex. V2.0. Scotts Valley, CA: Borland International, Inc; 1989
- ↵SPSS. Statistical Package for the Social Sciences. Release 6.0.1. Chicago, IL: SPSS Inc; 1993
- Streetly A,
- Grant C,
- Bickler G,
- Eldridge P,
- Bird S,
- Griffiths W
- ↵Schoenberg D, Klett M. Screening for congenital hypothyroidism in the Federal Republic of Germany. Past, present, and future. In: Therrel BL, ed. Advances in Neonatal Screening. New York, NY: Elsevier Science Publishers; 1987:25–29
- ↵Connelly J. Australian experience in neonatal thyroid screening, 1977–1985. In: Therrel BL, ed. Advances in Neonatal Screening. New York, NY: Elsevier Science Publishers; 1987:31–34
- ↵Irie M, Nakajima H, Inomata H, Noruse H, Suwa S, Takasugi N. Screening of neonatal hypothyroidism in Japan. In: Therrel BL, ed. Advances in Neonatal Screening. New York, NY: Elsevier Science Publishers; 1987:41–47
- ↵Italian Committee of Experts to Neonatal Mass Screening. Neonatal mass screening for metabolic disorders in Italy. Technical report of Italian Society of Pediatrics. In: Therrel BL, ed. Advances in Neonatal Screening. New York, NY: Elsevier Science Publishers; 1987:473–474
- Holtzman C,
- Stazyk W,
- Cordero J,
- Hannon H
- Sinai LN,
- Kim SC,
- Casey R,
- Pinto-Martin JA
- Copyright © 1998 American Academy of Pediatrics