PREDICTING THE C282Y HOMOZYGOTE GENOTYPE FROM SERUM FERRITIN AND TRANSFERRIN SATURATION IN YOUR PATIENT

A Lim; M Speechley PHD. Dept of Epidemiology and Biostatistics, University of Western Ontario, London, Ontario; PC Adams MD. Division of Gastroenterology, University Hospital, London, Ontario

Background: Serum ferritin and transferrin saturation have been used as a clinical guide to informally lead physicians to a diagnosis of Hemochromatosis. One etiology of Hemochromatosis is from C282Y mutations of the HFE gene on both alleles. In this study, a large population based sample of caucasians was used to create an equation to predict the C282Y homozygote genotype using serum ferritin and transferrin saturation levels. In addition, a smartphone application was developed to provide physicians with easy access to predicting C282Y homozygosity.

Methods: The HEIRS study screened over 100,000 participants for the C282Y mutation of the HFE gene. Analysis of the dataset (n=44,808) was done using a Bernoulli regression with logit link in the form where the binary response variable was C282Y homozygotes, and explanatory variables were serum ferritin, and transferrin saturations. The numerical solution of the Bernoulli regression generated a C282Y homozygotes probability equation for all serum ferritins, and transferrin saturations. A new Android application was developed using the Java computer language to allow for immediate C282Y homozygote probabilities to be estimated.

Results: An example of the analysis is a caucasian man with a transferrin saturation of 85% and a ferritin of 700 ng/L has a 32.2% [95% Confidence Interval 8.33-70.9%] probability of being a C282Y homozygote.

Conclusions: A large population based sample of 44,808 participants has led to the development of a new computer/smartphone based tool to predict the probability of C282Y homozygosity on your smartphone.

Analysis

A Bernoulli regression was used to analyze the HIERS data. The form of the Bernoulli regression was where the binary response variable was C282Y homozygotes, and explanatory variables were serum ferritin, and transferrin saturations (C282Y ~ Transferrin Saturation + Ferritin). The solution to the Bernoulli regression was preformed by R Statistical Software.

Males and Females were analyzed separately. Additionally, only caucasians were analyzed, therefore results only apply to caucasians.

Results

Males

The solution to the Bernoulli regression for Males (N=17323) is shown below.

The fit of the Bernoulli regression to the HEIRS data was assessed using Nagelkerke's Pseudo R-Squared. The value was calculated to be 0.440. A perfect fit to the data would be represented with a Nagelkerke Pseudo R-Squared value of 1.00.

The receiver operating characteristic curve (ROC) demenstrates the relationship between sensitivity, and specificity of this test. A perfect ROC would have an area under the curve of 1.0, this situation represents a test that has 100% specificity, and 100% sensitivity.

In this study, the Bernoulli regression for males was claculated using R Statistics and has a ROC with an area of is 0.908.

To the left is the graphical solution of the Bernoulli regression. It represents the probability of a C282Y Homozygote for every Ferritin and Transferrin Saturation. Additionally, there are the curves which represent the upper and lower confidence intervals (2.5%, and 97.5% confidence).

Females

The solution to the Bernoulli regression for Females (N=27485) is shown below.

The fit of the Bernoulli regression to the HEIRS data was assessed using Nagelkerke's Pseudo R-Squared. The value was calculated to be 0.343. A perfect fit to the data would be represented with a Nagelkerke Pseudo R-Squared value of 1.00.

In this study, the Bernoulli regression for females was claculated using R Statistics and has a ROC with an area of is 0.887.

To the left is the graphical solution of the Bernoulli regression. It represents the probability of a C282Y Homozygote for every Ferritin and Transferrin Saturation. Additionally, there are the curves which represent the upper and lower confidence intervals (2.5%, and 97.5% confidence).

Discussion

A large population based sample of 44,808 participants has led to the development of a new computer/smartphone based tool to predict the probability of C282Y homozygosity on your smartphone, or any internet connected device.

Limitations to the Bernoulli regression are that the Nagelkerke's Pseudo R-Squared values are 0.440 and 0.343 for male and female curves. The likely reason for these fit characteristics are that there are relatively few people in the dataset who are C282Y homozygotes. 137 of the 17,323 males screened are C282Y homozygotes, and 178 of the 27,485 females are C282Y Homozygotes.

The way we managed the poor Nagelkerke's Pseudo R-Squared values was by providing the 95% confidence values. This will help the end-user understand the limitations caused by the low volume of C282Y Homozygotes.