Trends in Childhood Anemia in a Massachusetts Health Maintenance Organization, 1987-2001

Abstract

Context

An important indicator of nutritional status within a pediatric population is the anemia prevalence rate. Limited national data are available regarding trends in anemia prevalence among non-low-income children.

Objective

To determine the prevalence of anemia over time among children enrolled in a health maintenance organization (HMO).

Design

Trend analysis, adjusted for clinical site, age group, sex, race/ethnicity, Medicaid insurance, and testing frequency within each year, and accounting for repeated observations of individual children across different years.

Setting

Massachusetts HMO.

Participants

72,729 children aged 6-59.9 months seen at 126,695 well-child visits from 1987 to 2001.

Main Outcome

Anemia prevalence by hemoglobin level.

Results

The observed prevalence of anemia was 8.9%, and was higher among younger children, blacks, boys, and those with Medicaid insurance. While the unadjusted anemia rate among children tested increased from 9.9% in 1987 to 11% in 2001, the proportion of children who were tested for anemia declined from 55% to 45% during the study period. After adjustment for demographics and testing frequency, the odds of anemia decreased over time (odds ratio 0.84, 95% confidence interval 0.76-0.93, per decade). Predicted anemia prevalence among those tested decreased from 8.9% in 1987 to 7.1% in 2001.

Conclusions

Anemia was less prevalent in this HMO population than has been reported in low-income children. After adjustment for testing frequency, the odds of anemia declined over time. As testing practices change to target children at higher risk, trend analyses should account for differences in screening rates over time. The computerized medical records of large health systems may serve as a valuable tool for nutritional surveillance among non low-income populations.

Background

The anemia prevalence rate within a pediatric population can be an important indicator of nutritional status. Iron deficiency and resulting anemia can have many adverse effects on child health, including delayed psychomotor development, impaired cognitive function, and increased susceptibility to lead toxicity.[1,2] Following the recognition of high rates of iron-deficiency anemia in the 1960s, interventions such as iron supplementation of infant formula, increased amounts of bioavailable iron in infant foods, and the Special Supplemental Program for Women, Infants, and Children (WIC) have substantially reduced rates of childhood anemia in the United States.[3,4] Nevertheless, iron deficiency remains the most common nutritional deficiency in the United States.

Most large-scale surveillance data describing declines in anemia come from the Pediatric Nutrition Surveillance System (PedNSS), a program established by the Centers for Disease Control and Prevention (CDC) to monitor trends in childhood nutritional indicators.[5] PedNSS reports on data collected through federally funded maternal and child health programs. The majority (82%) of PedNSS records are from low-income, nutritionally at-risk children participating in WIC. Overall anemia rates in the PedNSS population remained as high as 13.3% among children younger than 5 years of age in 2001.[5] However, this rate represents a decline in anemia prevalence in past decades,[5,6] and is similar to declines reported by a few smaller studies of individual pediatric practices.[7,8] Only limited data are available regarding trends in anemia prevalence over time among non-low-income children, and annual data are not available.[4]

We established a nutrition surveillance system in an eastern Massachusetts health maintenance organization (HMO), modeled after the PedNSS, to track nutritional indicators each year among a large population of predominantly non-low-income children.[9] This paper describes the results of the anemia analyses. We hypothesized that while anemia prevalence in our population would be lower than in the PedNSS population, rates would also have declined over time.

Methods

Subjects and Setting

The study population consisted of children enrolled from 1987 through 2001 in what was originally a staff model not-for-profit HMO in eastern Massachusetts. After 1998, the 13 urban and suburban health centers of the HMO became a group practice accepting outside health insurance, but continued to serve a similar patient population and to use the same computerized medical record. This computerized medical record includes data from all clinical encounters at the 13 sites. However, 2 notable changes occurred after 1998: The computerized database no longer provided information on race/ethnicity for the majority of children, and it did not include children with Medicaid insurance. During 2001, these health centers had over 160,000 enrolled members, including almost 10,000 children under age 5.

To generate the HMO nutrition surveillance system database, we queried the electronic medical record system for all well-child visits. We defined well-child visits to be those with a diagnosis code indicating routine healthcare, such as “periodic health review” or “well-baby care,” as well as those visits at which height or length was measured and immunizations were administered. We obtained data on 414,012 visits by 92,674 individual children aged 6 months to less than 5 years from 1987 through 2001. Encounter records were linked by date of visit with a separate laboratory database, and subsequently were deidentified before being provided to study investigators. Thus, the Institutional Review Board determined that this study was exempt from review.

Health plan protocol directed that anemia screening be performed on a venous blood sample sent for a complete blood count, including both hemoglobin and hematocrit tests. However, if the phlebotomist was unable to obtain an adequate specimen, a finger-stick capillary sample was collected for a spun hematocrit only. A total of 2698 visits had a hematocrit but no hemoglobin result; the majority of these (85%) were prior to 1996. Because the different methods of hematocrit measurement may yield noncomparable results, we used hemoglobin level alone to define anemia. We excluded visits at which no hemoglobin was obtained, as well as those with an extreme hemoglobin value of < 5 g/dL or > 20 g/dL.[10] From the remaining 134,427 visits, we randomly selected 1 visit per child per year in order to prevent bias from oversampling children who were evaluated more than once a year, as has been done previously.[6] However, because we did not adjust for a child's history of anemia in prior years, we were not able to determine the testing frequency by anemia status. We thus used data from 126,695 visits by 72,729 individual children for analysis of anemia trends.

From the clinical record, we obtained the child's sex, age at the time of the visit, and whether he or she had Medicaid insurance. Race/ethnicity was recorded in the record for less than half of subjects overall (43%), using the categories white, black, Hispanic, American Indian/Alaska Native, Asian, and other. Because of relatively small numbers, we combined all identified groups except white and black into the category “other race/ethnicity.”

Testing Frequency

We ascertained the frequency of hemoglobin testing within each year by dividing the number of children in whom a hemoglobin test was obtained in a given year by the total number of children seen at well-child visits within that year. To identify characteristics associated with the likelihood of being tested, we performed multivariable logistic regression, using as an outcome whether or not each individual child was tested in a given year; as predictors we used child age, race/ethnicity, sex, Medicaid status, and clinic, and year of visit as a continuous variable. We then added interaction terms for visit year with each of the subject characteristics that were independent predictors of being tested (age, race/ethnicity, and Medicaid status) to determine whether the relative likelihood of being screened for each of these groups changed during the study period.

Anemia Definition

We defined anemia according to the 1998 CDC guidelines,[11] as hemoglobin < 11.0 g/dL for children aged 6-23.9 months and < 11.1 g/dL for children aged 24-59.9 months. We did not adjust for altitude because most of eastern Massachusetts is at or near sea level, with a mean elevation of 500 feet.[12] We determined the percentage of tested children who were anemic for the overall population and by age, sex, race/ethnicity, and Medicaid status.

Trend Analysis

To determine the prevalence of anemia over time, we calculated the yearly proportion with anemia of all tested children with valid hemoglobin results and within each demographic category. Next, we performed multivariable logistic regression to determine the change in anemia prevalence over time. Anemia was the outcome of interest, and year of visit was the primary predictor. We additionally adjusted for clinic site, age group, sex, race/ethnicity, Medicaid status, and testing frequency within each year. We accounted for repeated observations of individual children across different years using generalized estimating equations.[13]

We generated predicted anemia prevalence among those tested from the multivariate model. To do this, we applied a constant proportion of age, sex, race/ethnicity, clinic site, Medicaid enrollment, and testing frequency for all study years, using the proportion in each group over the entire study period. We allowed the year of visit to change. We also calculated predicted anemia prevalence accounting only for age, sex, race/ethnicity, and Medicaid status, but not testing frequency. The predicted prevalence is the inverse logit of summed log odds multiplied by the above covariate values.

We used SAS version 8.2 (SAS Institute; Cary, North Carolina) for all statistical procedures.

Results

Population Characteristics

Table 1 shows characteristics of the 72,729 subjects who were seen at 126,695 visits for routine well-child care at the HMO clinical sites from 1987 through 2001 and had a valid hemoglobin result. The relative proportion of black and white children appeared to remain stable throughout the study period (data not shown), although data on race/ethnicity were sparse after 1998. Among the overall cohort, 29% of children were identified as white and 11% black, while 57% of children did not have race recorded. Approximately 8% of children had Medicaid insurance. Overall, anemia was present at 8.9% of visits at which a hemoglobin level was obtained; prevalence by child characteristics is presented in Table 1.

Table 1.

Characteristics of 72,729 Children Aged 6-59.9 Months With Screening Hemoglobin Obtained, Who Were Seen at 126,695 Well-Child Visits at a Massachusetts HMO, 1987-2001

Characteristic	Number (Percent) of Visits	Anemia* Prevalence
Total	126,695	8.9%
Sex
Male	64,518 (51)	9.2%
Female	62,177 (49)	8.5%
Age
6-23.9 months	57,535 (45)	11.0%
24-59.9 months	69,160 (55)	7.1%
Race/ethnicity
Black	14,335 (11)	13.9%
Other	4270 (3)	7.9%
Missing	71,909 (57)	9.1%
White	36,181 (29)	6.6%
Medicaid insurance
Yes	10531 (8)	10.9%
Missing	16344 (13)	10.2%
No	99820 (79)	8.4%

^*Anemia prevalence by visit among children who were tested, defined as hemoglobin < 11.0 g/dL ages 6-23.9 months, or hemoglobin < 11.2 g/dL ages 24-59.9 months, based on 1998 CDC guidelines.[11]

Testing Frequency and Anemia Trend Over Time

On unadjusted analysis, anemia prevalence among children screened appeared to increase over time from 9.9% in 1987 to 11% in 2001 (Figure). On bivariate analysis, the odds ratio of having anemia was 1.12 (95% confidence interval [CI] 1.06-1.17) for each additional study decade.

Hemoglobin tests were performed on 50% of children seen. Those children who had a hemoglobin test performed tended to be in demographic groups at higher risk for anemia: younger, non-white, and Medicaid-insured (Table 2). Boys and girls were tested with equal frequency (50%).

Table 2.

Factors Associated With Having a Hemoglobin Test Performed Among the 92,674 Children Seen for Well-Child Care at a Massachusetts HMO, 1987-2001

Characteristic	Frequency of Testing	Odds of Being Tested (95% CI)*
Sex
Male	50%	1.00 (0.99-1.02)
Female	50%	1.0 (referent)
Age
6-23.9 months	52%	1.25 (1.23-1.27)
24-59.9 months	48%	1.0 (referent)
Race/ethnicity
Black	61%	1.50 (1.45-1.55)
Other	58%	1.47 (1.39-1.54)
Missing	50%	1.15 (1.13-1.18)
White	45%	1.0 (referent)
Medicaid insurance
Yes	58%	1.21 (1.17-1.25)
Missing	48%	0.99 (0.97-1.02)
No	50%	1.0 (referent)
Year
Per decade	See text	0.81 (0.79-0.83)

^*Odds ratios are from multivariable analysis adjusting for all characteristics in the table as well as clinic site, and accounting for repeated observations of individuals across years.

Testing frequency was higher in the early part of the study period, peaking in 1991 when 55% of children seen had a hemoglobin test obtained, and dropping to a low of 45% in 1998. The proportion of children who had a hemoglobin test obtained within each year predicted the likelihood of anemia diagnosis. For each additional 10% of children tested in a year, the unadjusted odds ratio for anemia was 0.83 (95% CI 0.77-0.89). Thus, the fewer tests performed overall, the more likely an anemia diagnosis would be made for those who were tested. Additionally, during the study period, black children were increasingly more likely to receive an anemia test than were white children, and children with Medicaid were increasingly more likely be tested than were those without Medicaid (data not shown). These trends were apparent even when limiting the analysis to the period 1987-1998, when we had more complete information on race/ethnicity.

Given this variability in testing frequency across time, we examined anemia trends after adjusting for the proportion of children tested in each year. After adjustment for testing frequency alone, the odds of anemia declined over time, with an odds ratio of 0.90 (95% CI 0.83-0.98) per additional decade. In contrast, adjustment for subject characteristics alone (age, sex, race/ethnicity, and Medicaid status) did not result in an apparent decline in anemia over time, with an odds ratio of 1.04 (95% CI 0.97-1.12) per decade. The predicted yearly ascertained anemia from this model is shown in the Figure (dashed line).

After adjustment for testing frequency as well as demographics, study site, and clustering by individual, the odds of anemia declined over time. For each additional decade, the adjusted odds ratio of anemia was 0.84 (95% CI 0.75-0.93) (Table 3 and Figure). Male sex, younger age, Medicaid insurance, and black race/ethnicity continued to be associated with increased odds of anemia. We generated predicted anemia prevalence within each study year by performing multivariable regression and assuming a constant proportion of age, sex, race/ethnicity, clinic site, Medicaid enrollment, and testing frequency within each year of study. The predicted anemia prevalence decreased from 8.9% in 1987 to 7.1% in 2001 (Figure, solid line).

Table 3.

Adjusted Odds of Having Anemia According to Subject Characteristics and Year of Testing, Among 72,729 Children Seen at 126,695 Well-Child Visits at Which Hemoglobin Testing Was Performed

Variable	Odds of Anemia (95% CI)
Sex
Male	1.07 (1.02-1.12)
Female	1.0 (referent)
Age
6-23.9 months	1.68 (1.62-1.75)
24-59.9 months	1.0 (referent)
Race/ethnicity
Black	2.09 (1.93-2.26)
Other	1.13 (0.98-1.29)
Missing	1.23 (1.16-1.31)
White	1.0 (referent)
Medicaid insurance
Yes	1.10 (1.02-1.19)
Missing	1.13 (1.05-1.22)
No	1.0 (referent)
Time
Per decade	0.84 (0.76-0.93)
Testing frequency
Per 10% of children tested	0.82 (0.73-0.93)

Results are from multivariable logistic regression, adjusted for all characteristics included in the table as well as clinical site (13 sites), and accounting for repeated observations of individual children across years.

Because of the change in the computer medical record system after 1998, with race/ethnicity recorded for few children and no children with Medicaid included, we investigated whether the missing covariate information could have biased our results. Therefore, we reran our analyses using several different models: excluding the terms for race/ethnicity and Medicaid status in analysis of the full dataset; including only children without Medicaid; and including only the years 1987-1998 (Table 4). We found similar reduced odds of anemia over time in each of these models. In fact, after excluding the years 1999-2001 or excluding children with Medicaid, the decline over time appeared even stronger. These results suggest that missing information on subject characteristics did not cause the decline in anemia prevalence over time.

Table 4.

Additional Models Evaluating the Odds of Anemia Over Time, to Confirm That the Observed Effect Did Not Result From Missing Data on Medicaid and Race/Ethnicity After 1998

Dataset	# Children / # Visits	Covariates Included	OR (95% CI) per Additional Decade
All children 1987-2001	72,729 / 126,695	All	0.84 (0.76-0.93)
		All except race and Medicaid	0.89 (0.81-0.97)
Children without Medicaid 1987-2001	58,945 / 99,820	All except Medicaid	0.79 (0.71-0.88)
		All except race and Medicaid	0.77 (0.70-0.86)
All children 1987-1998	53,556 / 109,186	All	0.80 (0.71-0.88)
		All except race and Medicaid	0.79 (0.72-0.88)

Results from multivariable logistic regression, accounting for repeated observations of individual children across different years. Full models include as covariates age group, race/ethnicity, sex, Medicaid insurance, clinic site, visit year, testing frequency within each year.

Discussion

In this analysis of data from 72,729 children enrolled in an HMO, we observed declining odds of anemia over a 15-year period. Anemia rates were higher among males, younger children, blacks, and those with Medicaid insurance, and testing frequency was an important predictor of observed anemia prevalence. This study represents the first large-scale, surveillance program reporting anemia trends among a predominantly non low-income population of young children.

The frequency of anemia among children who were screened with hemoglobin tests was lower in this HMO cohort than in the nationwide and Massachusetts PedNSS data populations, which included only low-income children, most at nutritional risk.[14] In 2002 in Massachusetts, 13.7% of children younger than 5 years who were enrolled in WIC were anemic by hemoglobin or hematocrit, compared with 18% in 1995.[15] In the nationwide PedNSS population, the prevalence of anemia has declined similarly from 15.9% in 1995 to 13.3% in 2001.[5] Our estimated anemia prevalence of 7.4% in 2001 is lower than the reported prevalence among low-income populations, and is similar to the recently estimated national rate of 8.1% among all US children.[5,16]

We do not know the exact criteria used by clinicians to decide which children to refer for anemia screening. The observed decline in screening frequency during the study period probably resulted from a reduction in the frequency of screening all children as well as from a selective reduction in screening lower-risk children. If only a nonselective reduction in overall screening frequency occurred without any change in the risk for anemia among children screened, adjustment for screening frequency would not have affected the odds of anemia over time. If, however, clinicians became more selective in their decisions about whom to test, targeting those children at higher risk of being anemic, the observed anemia prevalence among those tested would have appeared to be higher than the true overall prevalence. In this case, adjustment for testing frequency would have resulted in estimates of anemia prevalence lower than those observed, as we found in the current analysis. In support of this assertion, children in this cohort who were black or on Medicaid were increasingly likely to be screened over the study period compared with whites or those without Medicaid.

Selective screening is not an issue in a planned survey such as the National Health and Nutrition Examination Survey, in which all children enrolled are tested. In PedNSS, all children are screened for anemia between 6 and 12 months of age. In the current study, statistical adjustment for testing frequency probably did not capture the subtleties of changing practice over time. However, without detailed information about the process by which screening decisions were made, adjustment for testing frequency probably accounts for some of the variation in screening practices. We believe that surveillance studies in clinical settings and other cohorts without universal anemia screening should account for such changes in testing frequency over time, to avoid misinterpretation of observed trends.

The decrease in screening frequency over time in this HMO setting is in accordance with recommendations for selective screening released by several advisory bodies. In 1998, the American Academy of Pediatrics (AAP) recommended that infants be screened for anemia by measuring hemoglobin or hematocrit between ages 9 and 12 months, and repeating the test 6 months later.[17] The AAP advised selective screening of those at high risk only in populations with overall rates of anemia below 5%.[17,18] The CDC (in 1998) and US Preventive Services Task Force (in 1996) have also advocated selective screening among populations at low risk for anemia, although their specific recommendations and screening thresholds differ.[11,19] In the study cohort, only 55% of children under age 5 who were seen for well-child care were screened a decade ago, and even fewer in recent years. Clinicians did appear to target children at higher demographic risk for anemia. We did not observe additional declines in testing frequency after 1998, the year in which the CDC and AAP published their recommendations for selective screening.

In addition to the issue of changing screening practices, the use of clinical records from an HMO or other health provider for nutritional surveillance may pose several challenges. The demographics of the patient population served may shift either suddenly or gradually. Individual children may enter or leave the system in any given year, and may be screened in another setting in that same year. Additionally, as in the present case, changes in record-keeping practices may influence data quality. Nevertheless, we believe that the computerized medical records from health systems can serve as a useful and relatively economical resource for tracking trends in anemia, and perhaps other health and disease indicators, among large populations of children.

The present study has several additional limitations. Although we extracted data solely from well-child visits, intending to obtain only those hemoglobin tests sent specifically for anemia screening, some blood counts may have been performed for other purposes or on children who were ill at the time of the scheduled well-child visit. Acute illness, even when mild, can transiently lower hemoglobin levels.[20] To be consistent with the PedNSS approach, we analyzed anemia rates according to prevalence, rather than using the absolute hemoglobin result. With this approach we were not able to determine whether hemoglobin levels became lower or higher among children with levels below the threshold for anemia. All subjects resided in Massachusetts, and our findings may not be generalizable to other populations.

Race/ethnicity was not recorded for approximately half of the children even prior to 1999, and thus we may not have completely adjusted for this strong predictor of anemia risk. As African- and Asian-American children are at higher risk for hereditary anemias, a change in the ethnic makeup of the population might have resulted in a change in anemia prevalence that did not result from more or less iron deficiency. However, we do not believe that the racial/ethnic makeup of the HMO patient population varied substantially over time. Thus, as discussed above, we do not believe that the missing information on race/ethnicity or the exclusion of children enrolled in Medicaid after 1998 from this cohort caused misleading results for the trend analysis.

Iron deficiency without anemia may also cause lasting harm.[21] However, not all anemia in young children results from iron deficiency, particularly as iron deficiency becomes less common. Although hemoglobin measurement is well standardized, relatively inexpensive, and easily performed in most settings,[11] other screening tests, such as ferritin or red blood cell indices, may be more specific for iron deficiency.[22,23] In the most recent National Health and Nutrition Examination Survey, iron deficiency, defined as abnormal results on 2 out of 3 tests of iron sufficiency, was present in 9% of children aged 1-2 years and 3% of children aged 2-5 years.[4] However, iron-deficiency anemia was present in only 3% of younger children and < 1% of older children, whereas anemia prevalence based on low hemoglobin alone was higher, approximately 5% to 11% depending upon age.[4] Because tests of iron status were infrequently performed in this HMO, we were unable to identify the overall prevalence of iron deficiency or iron-deficiency anemia.

In conclusion, overall anemia prevalence was lower in this HMO cohort than has been reported in low-income children, but showed similar declines over the past 15 years. Testing frequency was an important and time-varying predictor of apparent anemia prevalence that should be considered in future surveillance studies. In spite of the success of clinical and public health efforts to reduce anemia, anemia rates remained high in younger children and in blacks even in this predominantly non-low-income population. In these higher-risk populations, hemoglobin levels remain an appropriate screening test for iron deficiency. Pediatric providers should continue to screen and treat children at higher risk to avoid the “preventable tragedy”[24,25] of permanent developmental sequelae from iron-deficiency anemia.

Figure.

Observed and adjusted anemia prevalence trends in a Massachusetts health maintenance organization from 1987 through 2001.

The observed proportion of positive tests for anemia (gray line) among children screened increased over time. After adjustment for age, sex, Medicaid status, and race/ethnicity (dashed line), the odds of anemia still showed a slight increase over time (odds ratio 1.04 [95% CI 0.97-1.12] per decade). However, after adjustment for the frequency of testing within each year as well as subject demographics (solid line), the odds of anemia decreased (odds ratio 0.84 [95% CI 0.75-0.93] per decade). We generated predicted ascertained anemia prevalence by assuming a constant proportion of covariates across years. Data are from 126,695 visits by children aged 6-59.9 months seen for well-child care at a Massachusetts HMO from 1987 through 2001.

Funding Information

The research in this article was supported by grants from the National Institutes of Health (HL 68041, HD44807), training grants (T32 PE 11011-15), the Centers for Disease Control and Prevention (through task order #0957-007), and Harvard Medical School and the Harvard Pilgrim Health Care Foundation.