Wunderink et al1 have systematically analyzed the clinical factors associated with the development of BK infection in a cohort of 407 living donor kidney transplant recipients. 111 (27%) recipients in this cohort developed BK viremia and 12 (3%) developed BK nephropathy during the first posttransplant year. In this carefully done study the authors show that the recipient HLA-B51 status correlates with a lower risk of BK viremia. 36 of the recipients were HLA-B51 positive and only 2 developed BK viremia (P = 0.002). The HR for developing BK viremia in the HLA-B51 positive recipients was 0.18 (95%: 0.04–0.73). The authors suggest that knowing the HLA-B51 status may help stratify recipients for the risk of BK infection during the posttransplant period.
In the multivariate analysis, the donor BK seropositivity had a hazard ratio [HR] of 1.61 (95% confidence interval, 1.38–1.88) and the recipient HLA-B51 status an HR of 0.18 (95% confidence interval, 0.04–0.73). The analysis showed that the donor BK seroreactivity was the strongest association with BK viremia.
The authors also used published databases of the amino acid sequence of BK proteins and compared these with the published amino acid sequences of the HLA-B51 protein and show evidence of compatible sequences. This analysis showed 4 viral protein peptide sequences which might be predicted to anchor specificity with the HLA-B51 cytotoxic T lymphocyte epitopes. They suggest that this finding offers a plausible biologic explanation for the relevance of the recipient HLA-B51 status as a protective mechanism.
These unique and provocative observations are important; however, these results should be considered as preliminary. Although this analysis shows an association of the HLA-B51 status with lower risk of BK infection, this should be interpreted cautiously. An important statistical concern is the problem of multiple hypothesis testing and the risk of type I error (false positive). It is a well-established statistic concept that multiple hypothesis testing increases the risk of a random false positive test (type I error). The greater the number of hypothesis tested on the same data set the higher the risk of false positive. There are several statistical techniques uses to correct for multiple testing. The most common technique is the Bonferroni correction. A marginally more conservative correction can be obtained by using the Šidák correction which was the method used in the current study. The P values were corrected in 2 ways; by the number of HLA alleles tested within each locus, and by the total number of tested HLA alleles. There were 78 HLA alleles tested. They determined the distribution of HLA genotype in the 407 viremic and non viremic recipients. For HLA-B51 this comparison showed the crude P value by Fisher exact test was 0.001, but when adjusted for the number of HLA locus the P value was 0.035, and after adjusting for the total number of HLA comparisons the P value was 0.093. Thus the association found once corrected was not statistically significant. While one should not place too much emphasis on a P value, the inherent variability in these types of studies does require one to look cautiously at only marginally significant events.2
A minor point lay within the multivariable analysis. If one uses a rule of thumb of 1 variable for every 10 events then 22 variables for the 111 events noted indicates that the model may be over fitted. In addition the point estimate used for HLA-B51 may be prone to large variability and thus given the rather small number of patients with this allele the hazard ratio may not be fully reproduced in a validation study.3
As the authors candidly point out in the discussion, another critical issue is the need for independent validation. Because the study was in retrospective design, the study variables may not have been established a priori. It will be necessary to validate the results in an independent cohort. In validation studies, the models’ performance is usually inferior to the performance in the discovery cohort. Before testing for HLA genotype can be used clinically to estimate the risk for BK infection, the performance characters of the test will need to be determined. This will require the construction of a receiver-operator curve with calculation of the area under the curve or the C-statistic for estimating the reliability of the test. This is an indicator of the positive and negative predictive value after taking into account the important impact of the prevalence of the condition.
Finally, there is the issue of biologic plausibility. Although the authors suggest that amino acid sequences in the BK virus has some analogy based on published database analysis of the amino acid sequences in the HLA B51 molecule exposed sequences, this remains speculative particularly as no data was given regarding whether other HLA alleles also had the same in vivo characteristics as B51. Further in vitro and animal model studies will need to be performed to better understand the biologic pathways involved in this complex process involving the host immune response to the BK viral infection and to better understand how the HLA phenotype may be involved in the process of viral clearance.
The authors should be congratulated for their high level of rigor and demonstrating this type of rigor does not mitigate the need for thorough and robust validation.
1. Wunderink H, Haasnoot GW, de Brouwer CS, et al. Reduced risk of BK polyomavirus infection in HLA-B51 positive kidney transplant recipients. Transplantation. 2018;103:604–612.
2. Lin YT, Lee WC. Importance of presenting the variability of the false discovery rate control. BMC Genet. 2015;16:97.
3. Singh SK, Kaplan B, Kim SJ. Multivariable regression models in clinical transplant research: principles and pitfalls. Transplantation. 2015;99:2451–2457.