Key messages
• The two most common tests (serum or plasma retinol and retinol-binding protein) cannot reliably identify at-risk individuals with or without vitamin A deficiency in the community or clinics.
• This may mean that using these tests could lead to over- or underestimated levels of low vitamin A at the population level and incorrect diagnosis of individuals.
• Estimates of test accuracy should improve as more studies become available.
Why is it important to improve screening for vitamin A?
Vitamin A, also called 'retinol', is an essential nutrient. It is found in liver, eggs, dairy products, and most green leafy or orange fruit and vegetables. However, lack of vitamin A (vitamin A deficiency) is a common public health problem, especially in low-income countries. People with vitamin A deficiency may suffer from poor growth, an unhealthy immune system, night blindness and blindness. They are more likely to catch infectious diseases. Young children and pregnant women are most at risk.
In order to decide whether treatment with vitamin A supplements is needed or not, for individuals and communities, we need to know the existing levels of vitamin A deficiency. This requires accurate tests. Not recognising vitamin A deficiency when it is present may result in delayed or no vitamin A supplementation; vitamin A deficiency worsens, leading to scarring and vision problems. An incorrect diagnosis of vitamin A deficiency may mean that individuals are given vitamin A supplements when they are not needed, increasing the risk of vitamin A intake above recommended levels.
What tests measure vitamin A status?
The most accurate tests for vitamin A status (called the 'reference standards’) are liver biopsy and retinol isotope dilution. Liver biopsy involves removing a small sample of someone's liver and examining it under a microscope in a laboratory. Retinol isotope dilution involves giving a small dose of vitamin A and then examining a blood sample. These tests are invasive, costly, time-consuming, and technically demanding.
Other tests to measure a person’s vitamin A status are less invasive, less expensive, and quicker, but may not be so accurate. They are called 'index tests'. They look for signs of vitamin A ('biomarkers') in a blood sample. The most common tests are 'serum or plasma retinol' (SR) and 'serum or plasma retinol binding protein' (RBP). A positive test result indicates vitamin A deficiency, while a negative test result indicates the absence of vitamin A deficiency.
What did we want to find out?
We wanted to determine the accuracy of index tests compared with the reference standards for detecting vitamin A deficiency in people at risk of vitamin A deficiency.
What did we do?
We searched for studies that investigated the accuracy of the index tests compared to a reference standard and combined the results of the studies.
What did we find?
We included 40 studies that compared at least one index test and one reference test. Studies took place in North, Central and South America; Europe; South and Southeast Asia; and Africa. Participants were aged from 1 month to over 80 years. The percentage of people with vitamin A deficiency averaged between 4% and 60%. The results below are based on a hypothetical group of 1000 people.
SR versus retinol isotope dilution (23 studies)
In 1000 people where 100 (10%) have vitamin A deficiency measured with retinol isotope dilution, 82 people would have a SR result indicating vitamin A deficiency is present. Of these, 72 would be incorrectly classified as having vitamin A deficiency. Of the 918 people with a result indicating that vitamin A deficiency is not present, 90 would be incorrectly classified as not having vitamin A deficiency.
SR versus liver vitamin A (16 studies)
In 1000 people where 100 (10%) have vitamin A deficiency measured with liver vitamin A, 206 people would have a SR result indicating vitamin A deficiency is present. Of these, 153 would be incorrectly classified as having vitamin A deficiency. Of the 794 people with a result indicating that vitamin A deficiency is not present, 47 would be incorrectly classified as not having vitamin A deficiency.
RBP versus retinol isotope dilution (8 studies)
In 1000 people where 100 (10%) have vitamin A deficiency measured with retinol isotope dilution, 266 people would have a RBP result indicating vitamin A deficiency is present. Of these, 216 would be incorrectly classified as having vitamin A deficiency. Of the 734 people with a result indicating that vitamin A deficiency is not present, 50 would be incorrectly classified as not having vitamin A deficiency.
What are the limitations of the evidence?
Our confidence in the evidence is limited for several reasons. Most studies were small with little information about how they were done, so we couldn't judge how robust their methods were. Studies were not designed to determine the accuracy of index tests, so we need to be careful about interpreting their results. Finally, we didn't find information on all the tests.
How up to date is this review?
The evidence is up-to-date to August 2022.
Available data indicate that methods to determine vitamin A deficiency had generally low sensitivity, when estimable (0% to 54%), and generally high specificity (74% to 94%) in individuals at risk for vitamin A deficiency. Estimates should be interpreted with caution because no included studies were designed or conducted as DTA studies. Data assessing the accuracy of the breast milk vitamin A, RDR, and MRDR compared to reference standards, particularly in patients with vitamin A deficiency, are limited.
Vitamin A deficiency is a highly detrimental micronutrient deficiency associated with poor growth, impaired immune responses, increased incidence of disease, ocular impairments, and maternal and child mortality. Reliable diagnostic assessment of vitamin A status is crucial to inform its clinical management. Currently, direct index measures and dose response biomarkers have been developed to provide assessments of vitamin A status.
To determine the accuracy of index tests routinely used as markers of subclinical vitamin A deficiency in individuals at risk for vitamin A deficiency. Secondary objectives are to assess covariates as sources of heterogeneity for the accuracy of index tests routinely used as markers of subclinical vitamin A deficiency.
We searched CENTRAL, MEDLINE, Embase, and six other databases up to 18 August 2022, without restriction (any sex, age, pregnancy status, breastfeeding status, physiological condition, living in any country).
We included any studies implementing concurrent measurement of at least one reference standard and one index test to measure vitamin A status. Eligible studies included cross-sectional or cohort-accuracy studies; longitudinal studies; and direct, indirect, and random comparison studies, in which multiple index tests with a reference standard were evaluated. Interventional studies measuring vitamin A status following supplementation or intervention were also included, while case-control studies defining cases by vitamin A status were excluded.
Two review authors independently screened studies and extracted data. We evaluated the methodological quality, that is, risk of bias of included studies and their applicability by using the Quality Assessment of Diagnostic Accuracy Studies (QUADAS‐2) tool. When meta‐analysis was appropriate, we used random-effects bivariate models to obtain pooled estimates of sensitivity and specificity. We stratified all analyses by the reference standard and cutoff used, and assessed the certainty of the evidence using GRADE.
We included 40 studies reporting 65 records. None of the studies was designed as a diagnostic test accuracy (DTA) study, limiting our analyses and assessments. Index test performance was described by 25 studies for serum or plasma retinol (SR) versus retinol isotope dilution (RID), 16 studies for SR versus liver vitamin A, eight studies for retinol-binding protein (RBP) versus retinol isotope dilution (RID), three studies for RBP versus liver vitamin A, one study for breast milk vitamin A versus RID, three studies for relative dose response (RDR) versus RID, and four studies for RDR versus liver vitamin A. No studies evaluating modified RDR were eligible for inclusion. Specificity data were available from all studies; sensitivity was often estimable from only a portion of studies due to some studies having no condition-positive cases according to the reference standard (zero true positive and false negative cases). One comparison, RDR versus RID, yielded no sensitivity data, therefore we could evaluate only pooled specificity data. We generally judged risk of bias as 'unclear' across studies.
Serum or plasma retinol for diagnosing vitamin A deficiency
SR pooled sensitivity against RID at the 0.1 μmol/g cutoff was 10% (95% confidence interval (CI) 2 to 38; 23 studies, 385 participants; very low-certainty evidence), and specificity was 92% (95% CI 85 to 96; 23 studies, 1110 participants; low-certainty evidence).
SR pooled sensitivity against RID at the 0.07 μmol/g cutoff was 13% (95% CI 4 to 34; 24 studies, 246 participants; very low-certainty evidence), and specificity was 94% (95% CI 87 to 97, 24 studies, 1295 participants; low-certainty evidence).
SR pooled sensitivity against liver vitamin A at the 0.1 μmol/g cutoff was 53% (95% CI 30 to 75; 16 studies, 192 participants; very low-certainty evidence) and specificity was 83% (95% CI 63 to 93; 16 studies, 370 participants; moderate-certainty evidence).
SR pooled sensitivity against liver vitamin A at the 0.07 μmol/g cutoff was 54% (95% CI 33 to 74; 16 studies, 137 participants; very low-certainty evidence) and specificity was 79% (95% CI 57 to 91; 16 studies, 348 participants; moderate-certainty evidence).
Retinol-binding protein for diagnosing vitamin A deficiency
RBP pooled sensitivity against RID at the 0.1 μmol/g cutoff was 50% (95% CI 33 to 67; 8 studies, 30 participants; low-certainty evidence) and specificity was 76% (95% CI 72 to 81; 8 studies, 730 participants; moderate-certainty evidence).
RBP pooled sensitivity against RID at the 0.07 μmol/g cutoff was 45% (95% CI 31 to 59; 8 studies, 47 participants; low-certainty evidence) and specificity was 77% (95% CI 71 to 82; 8 studies, 711 participants; moderate-certainty evidence).
RBP pooled sensitivity against liver vitamin A at the 0.1 μmol/g cutoff was 0% (95% CI 0 to 100; 3 studies, 12 participants; very low-certainty evidence) and specificity was 98% (95% CI 84 to 100; 3 studies, 40 participants; very low-certainty evidence).
RBP pooled sensitivity against liver vitamin A at the 0.07 μmol/g cutoff was 0% (95% CI 0 to 100; 3 studies, 9 participants; very low-certainty evidence) and specificity was 98% (95% CI 85 to 100; 3 studies, 43 participants; very low-certainty evidence)
Relative dose response for diagnosing vitamin A deficiency
RDR pooled sensitivity against RID at the 0.1 μmol/g and 0.07 μmol/g cutoffs were not estimable due to lack of true-positive and false-negative cases from three studies. RDR pooled specificity against RID at the 0.1 μmol/g cutoff was 89% (95% CI 26 to 99; 3 studies, 34 participants; low-certainty evidence), and RDR pooled specificity against RID at the 0.07 μmol/g cutoff was 91% (95% CI 54 to 99; 3 studies, 34 participants; low-certainty evidence).