A panel convened by the CDC recommended that transferrin saturation be used as the initial test for hemochromatosis (1). Transferrin saturation can be calculated from the ratio of serum iron concentration to total iron-binding capacity (TIBC) expressed as a percentage [(serum iron/TIBC) × 100] or, instead, using the sum of serum iron and unsaturated iron-binding capacity (UIBC): TIBCcalc = serum iron + UIBC. We conducted a round-robin interlaboratory comparison study to assess the comparability of results for serum iron and iron-binding capacity assays.
To solicit participants, we sent invitations to iron experts (n = 10) and an announcement was placed on the Clinical Chemistry/General Topics listserv and in the American Association for Clinical Chemistry Nutrition Division newsletter; separate announcements were sent to at least 25 providers of US diagnostic services. A CDC Institutional Review Board approved the protocol (no. 2552). A contracted blood donor center collected 1 unit of blood from each anonymous human donor who gave informed consent. The donor center was also instructed to invite hemochromatosis patients to participate to ensure samples with increased iron status but free of iron supplement use. Serum was made by collecting the blood in bags with no anticoagulant. The blood was allowed to clot for several hours and centrifuged, and the serum was decanted. No manipulation (i.e., addition of analyte) of serum samples was done. Each serum sample was tested for serum iron and TIBC in the Nutrition Laboratory at CDC, using the Alpkem Flow IV colorimetric methods (2)(3). The methods are semiautomated applications of the NCCLS-recommended manual methods (4). On the basis of the transferrin saturation values, we combined four samples to create a pool representing marginal iron status (low pool), three to create a pool that was borderline increased and therefore close to the proposed cutoffs for the diagnosis of iron overload (5) (medium pool), and five to create a pool representing iron overload (high pool).
Each laboratory was asked to analyze the three serum pools. The low and high pools were each analyzed over 3 days in 4 replicates each (2 replicate measurements per vial, 2 vials per pool per day) for a total of 12 replicate values. Because of the small size of the medium pool, the number of replicates was reduced. The medium pool was analyzed over 3 days in two replicates each (two replicate measurements per vial, only one vial per pool per day) for a total of six replicate values.
We obtained a total of 25 sets of results for serum iron, 13 for TIBC, and 16 for UIBC. All values reported by the laboratories were μg/dL. The values were converted to μmol/L by multiplying by 0.179. For iron determination, we a priori identified eight types of assays, which we labeled A–H as defined in Table 1⇓ . The various serum iron and iron-binding capacity measures, instrumentation, and analytic methods used by the 25 participating laboratories are available in the Data Supplement that accompanies the online version of this Technical Brief at http://www.clinchem.org/content/vol49/issue10/.
We tested for outliers using the Dixon method (6) and set the critical rejection value at 0.01. We then calculated the all-laboratory consensus mean and SD for each pool and determined whether each individual result was within 3 SD of the mean. Results outside of this consensus range were excluded from all analyses. Two serum iron results were excluded: one medium-pool result from laboratory 23 was excluded by use of the Dixon method (6), and one low-pool result from laboratory 19 was outside the consensus range. Before outlier detection for TIBC, we excluded medium-pool results from laboratory 17 because they noted a gelatinous material in two of the medium-pool vials; the material was removed, but the data from these specimens were anomalous for both 17a and 17b. One TIBC result was excluded, a low-pool result from laboratory 13 that was outside the consensus range. Laboratory 5 did not report raw UIBC data for the high pool and was excluded. Ten UIBC results were excluded: 4 low-pool results from laboratory 21 and all 6 medium- pool results from laboratory 16 were outside the consensus range.
Microsoft Excel and SAS, Ver, 8, were used for all analyses (7). For each laboratory, the mean, SD, and CV were computed. Among-method CVs for each pool were computed by dividing the SD of the seven method group means (each method group mean was computed as the mean of the means of laboratories using the same method) by the grand mean of the seven method group means. Within-method CVs for each method group were computed by dividing the SD of the laboratory means (for all laboratories using the same method) by the grand mean of these laboratory means.
To test for methodologic differences in serum iron measurements, we tested for the within-method heterogeneity of variance using Bartlett’s test (8). We found unequal method variances for all three serum iron pools. Therefore, using SAS GLM, we performed a weighted (weight = 1/variance) two-way ANOVA with method and laboratory (nested within method) as variables and laboratory treated as a random variable. Results were adjusted using the Bonferroni method (9). The same procedure was used to assess differences in TIBC and UIBC.
Of the 25 participating laboratories, 22 were in the US, and the other 3 were in Belgium, Sweden, and Wales. The among-laboratory and among-method means (SD) and among-run and within-run CVs for serum iron, TIBC, and UIBC measurements are listed in Table 1⇓ . The quartiles and minimum and maximum observations for the measurements by pool are shown in Fig. 1⇓ . The within-method variation (among-run and within-run) was <3% for serum iron and TIBC for all method types and pools. For UIBC, the within-method variation was also fairly low for all method types for the low pool; however, the within-method variation for the medium and high pools was between 5% and 34%.
When we grouped laboratories by serum iron assay method and performed a two-way ANOVA using Bonferroni adjustment, we found no statistically significant differences among the serum iron methods for the low (P = 0.89), medium (P = 0.16), or high pool (P = 0.40). Similarly, when we grouped laboratories by serum iron assay method we found no significant differences among the TIBC assays for the low (P = 0.93), medium (P = 0.38), or high pool (P = 0.66) or among the UIBC assays for the low (P = 0.25), medium (P = 0.67), or high pool (P = 0.81). Results (mean, minimum, and maximum observations) for iron and iron-binding capacity grouped by method type are presented as box plots in a figure in the online Data Supplement.
Our study indicates that there were no significant differences in results generated by methods currently in use to determine serum iron concentrations. We observed low across-laboratory and across-method variability for serum iron, TIBC, and UIBC.
We present SDs to compare across-laboratory and across-method variability rather than CVs. This was done to assure comparability between TIBC and UIBC. When the transferrin protein becomes highly saturated with iron, the mean UIBC becomes numerically quite small; therefore, its CV looks high compared with that of TIBC, although their SDs are fairly comparable.
Although the TIBC results for the medium pool were somewhat varied (37.6–60.8 μmol/L), analysis of the raw data shows that the values for each laboratory were consistent from one vial to the next and from one day to the next. This is true for the laboratories producing both the lowest results and the highest results. If the medium pool had been lacking homogeneity, it should have manifested itself in a large vial-to-vial variability. Laboratories 1 (method A) and 9 (method E), which both used copper correction and true protein removal, measured much lower concentrations in this pool than the other laboratories did. Perhaps some naturally occurring protein in one of the donors or some drug or other interfering substance, such as a non-protein iron-binding substance, in the medium pool was successfully removed by these methods but not by the other methods. Four of the TIBC results from the other methods were >50 μmol/L. When we excluded the two with mean TIBC results >53 μmol/L, the overall mean for the medium pool decreased to 43.2 μmol/L. We are unaware of laboratory circumstances that could have led to these higher values.
Results from one or two laboratories had a discernible impact on UIBC. For the medium pool, the mean UIBC values ranged from 9.3 to 16.8 μmol/L; the majority of laboratories had values in the range of 12.5–14.3 μmol/L. When we excluded the results from laboratory 8 [9.3 (2.0) μmol/L], the among-laboratory mean for the medium UIBC pool increased to 15.0 (1.2) μmol/L. For the high pool, the mean UIBC values ranged from 0.9 to 5.5 μmol/L; the majority of laboratories had values in the range of 1.8–3.6 μmol/L. When we excluded the results from laboratories 7 and 15 because of their relatively low means and high SDs, the overall among-laboratory mean (SD) for the high pool increased to 3.5 (0.9) μmol/L.
There are additional limitations to our study. We had few reference laboratories, and there are no accepted reference methods for TIBC or UIBC. Although the performance of the laboratories that we studied may not be representative of the performance of clinical laboratories in general, we believe this not to be the case. When the College of American Pathologists added TIBC to their Chemistry Survey (in 1997 on a trial basis and formally in 2002), their all-method-principle and all-instruments CVs ranged from ∼4% to 7% (10)(11), similar to our findings. We chose to exclude results from two laboratories that reported a gelatinous material in the TIBC medium pool and chose not to impute values for one laboratory that did not report raw values for the UIBC low pool. Additionally, we excluded 13 values from the ∼1650 analyte values; these, however, represent <1% of all values and should not reduce interpretation of analytic validity.
We gratefully recognize those involved in the envisioning of the project, including Barbara A. Bowman, Mary K. Serdula, Mary E. Cogswell, Larry Grummer-Strawn, Dan Huff, and Gary L. Myers, as well as all of the directors and personnel of the following laboratories: Ortho Clinical Diagnostics (Rochester, NY), Collaborative Laboratory Services (Ottumwa, IA), Fairview University Medical Center (Minneapolis, MN), Quest Diagnostics (Teterboro, NJ; Tucker, GA; Dublin, CA; Van Nuys, CA; Auburn Hills, MI; St. Louis, MO; Wood Dale, IL), Sonora Quest (Tempe, AZ), ARUP Laboratories (Salt Lake City, UT), William Beaumont Hospital (Royal Oak, MI), Atlanta VA Medical Center (Atlanta, GA), Emory University Medical Laboratories (Atlanta, GA), Spectra East Laboratory (Rockleigh, NH), Reference Diagnostics (Bedford, MA), South CA Permanente (North Hollywood, CA), University of Wisconsin (Madison, WI), Rochester General Hospital Laboratory (Rochester, NY), Harborview Medical Center (Seattle, WA), University Hospital of Wales (Cardiff, UK), UZ-K.U. Leuven Medical Center (Leuven, Belgium), and Sahlgrenska University Hospital (Goeteborg, Sweden).
- © 2003 The American Association for Clinical Chemistry