Study population
A cohort study was designed for our research purpose. Patient data were collected from a large database set up in the laboratory of the Albert Schweitzer Hospital, Dordrecht, the Netherlands. This project was of multidisciplinary origin; general practitioners (GPs), internal medicine physicians and clinical chemists participated. Eighty-one of the 150 GPs in the city of Dordrecht area, the Netherlands, agreed to participate in this project, which started on 1 February 2007. The department of clinical chemistry of the Albert Schweitzer Hospital is the referral laboratory for GPs in the area.
If a participating GP requested a blood test for a patient aged ≥ 50 years and this revealed a low hemoglobin concentration according to the references values of the participating laboratory (i.e. hemoglobin <13.7 g/dL and <12.1 g/dL for males and females, respectively), a further extensive laboratory assessment was performed. This assessment consisted of measuring hemoglobin, mean corpuscular volume (MCV), reticulocyte count, leukocyte count, thrombocyte count, lactate dehydrogenase, vitamin B12, folic acid, creatinine, ferritin, transferrin, and serum iron. These laboratory tests were chosen based on the Dutch guideline and tests regularly ordered by the participating internal medicine physicians [11]. The test results, together with the patient’s sex and age were stored anonymously in the database. As we wished to study anemia etiologies during diagnostic work-up, we excluded data of patients already known to have anemia in the previous two years.
This study was approved by the institutional review board of the Albert Schweitzer Hospital and was conducted in accordance with the Declaration of Helsinki.
Definitions
We defined a set of 8 laboratory-orientated etiologies of anemia based on literature, the Dutch general practitioners’ guideline of anemia and the references values of the participating laboratory (table 1) [3 – 6, 11 – 14]. These eight etiologies were ACD, renal anemia, IDA, hemoglobinopathy, suspected hemolysis, suspected bone marrow disease, vitamin B12 deficiency and folic acid deficiency. The use of these definitions ensures a uniform representation of the laboratory etiologies and takes into account the possibility of multiple etiologies in a patient. The strict application of the definitions made the combination of IDA and ACD impossible.
To be able to visualize and analyze trends between anemia etiology and patient characteristics, we clustered some etiologies that had a low incidence. Thus, vitamin B12 deficiency and folic acid deficiency were taken together, and hemoglobinopathy, suspected hemolysis and suspected bone marrow disease were clustered as ’other etiologies’. In addition, hemoglobin values were clustered into three groups proportionally. Severe anemia was defined as hemoglobin ≤ 9.7 g/dL; moderate anemia as hemoglobin > 9.7 and ≤ 12.9 g/dL for males and > 9.7 and ≤ 11.3 g/dL for females. Mild anemia was defined as hemoglobin > 12.9 and < 13.7 g/dL for males and > 11.3 and < 12.1 g/dL for females.
Statistical analysis
Laboratory protocol violations had resulted in missing laboratory values (table 1). To avoid selection bias as a consequence of the exclusion of patients with missing values, we applied single imputation using an expectation-maximization algorithm [15]. For nine laboratory values, the percentage of missing values was below one. For creatinine, however, it was 19.6%. This could be ascribed to the fact that GPs could follow two pathways when they requested laboratory analysis, one of which did not include creatinine. Still, this percentage was low enough to allow single imputation to be applied. Characteristics of the study population are described in terms of frequency, median and interquartile ranges. The distribution of underlying etiologies of anemia and the multiple aspect of the etiologies is described using counts and relative frequencies. We visualized the distribution of the etiologies of anemia by age, sex and level of anemia severity. To assess whether the etiology of anemia is associated with age, sex and/or the severity of anemia, we performed a multinomial logistic regression analysis. The dependent variable was anemia etiology (ACD, renal, IDA, vitamin B12 or folic acid deficiency, other, multiple etiologies or uncertain). The independent variables were age (continuous variable), sex (male/female) and anemia severity (mild/moderate/severe). Model assumptions were tested and met. We used SPSS for Windows, version 24 (IBM Corp., Armonk, NY, USA), for data handling and analyses. All statistical tests were two-sided and we considered a p-value p < 0.05 statistically significant.