Strengths and limitations
The study has several strengths. Compared with previous, small, studies conducted with patients with clinically high risk of HZ, this large study of a general population provides greater statistical power and generalisability. In addition, the vitamin D levels were measured systematically, and the proportion of covariate missingness was relatively low. The linkage between UK Biobank and the primary and secondary care records also enabled participants to be followed up for a long time and incident cases to be identified.
Nevertheless, some limitations also need to be acknowledged. The exposure and some covariates are likely to be time dependent, but measures used were taken at baseline. In the authors’ previous analysis, the proportion of vitamin D deficiency was lowest in summer and more prevalent in winter and spring;19 in another study measuring vitamin D repeatedly, the intraclass correlation coefficient between two vitamin D measurements after 5 years was only 0.59, which was moderately reliable.21 In the analysis presented here, Weibull regression was used, which assumes hazards increase during follow-up, and adjustments were made for vitamin D testing seasons in the model to minimise the effect of seasonal variation. In the sensitivity analysis, Cox model regression was used to adjust for potentially time-varying hazards and the results remained similar. Nevertheless, vitamin D status would change over time, which may introduce exposure misclassification.
It should also be noted that, despite the completeness of most covariates, more than half of the data were missing for self-reported vitamin D supplementation; as such, this variable may not reflect the real vitamin D supplementation use, and its association with the outcome needs to be interpreted with caution. A positive trend of association between GP-prescribed vitamin D supplementation and HZ was noted in the crude and partially adjusted models in the sensitivity analysis. This association may be due to confounding by indication, as well as the underestimation of unreported food fortification. People receive vitamin D prescriptions to prevent or treat vitamin D deficiency, but that indication for the prescription was not considered in the study presented here. Vitamin D food fortification is another main source of vitamin D supplementation in the UK primary care setting;22 however, because of the limitation of data availability, it was not included in the analysis.
The outcomes may also be underascertained. HZ was defined using EHRs, but people with a greater number of comorbidities may visit their primary care physicians more frequently than people without underlying chronic diseases; as such, HZ in these people is more likely to be diagnosed, whereas mild shingles among a younger or healthier population might not be noticed.23 In the study population, although the proportions of people with certain conditions were slightly higher in those who were deficient in vitamin D, the overall distributions of comorbidities and immunosuppression were similar across different vitamin D statuses. Any ascertainment bias in the study presented here should be non-differential.
Misclassification of HZ outcomes cannot be ruled out. Studies using EHRs assume that individuals have a disease if they have the corresponding diagnostic codes; conversely, people without specific diagnostic codes are assumed not to have the disease. It is possible that some participants with HZ did not visit their GP or did not have a confirmed diagnosis, so their disease statuses might have gone unrecorded.
Regarding possible ascertainment bias, a study in the US reported that using the ICD-9 code for definite or possible herpes virus could identify 98% of HZ cases (sensitivity 98%), and the positive predictive value was also very high (PPV 93%).24 Furthermore, the financial barrier to access health care in the UK is generally lower than in the US, which may increase the sensitivity of HZ diagnoses in the EHR databases.
Residual confounding effects also cannot be ruled out. Using diagnostic codes from the linked records may underestimate the true prevalence of some diseases, such as CKD. In studies using EHRs, serum creatinine levels are more often used to diagnose CKD instead of using diagnostic codes alone;25 however, laboratory test results are not available in the linked EHRs of UK Biobank. To enhance the sensitivity of detecting comorbidities, the authors of the study presented here included self-reported, non-cancer health conditions in the analysis, but the overall prevalence of CKD was still much lower than the national prevalence during the same period.26
Comparison with existing literature
This is, to the authors’ knowledge, the first published study assessing the association between vitamin D status and incident HZ in the general population. Previous studies on the topic have been conducted among people with immunosuppression; as an example, a case–control study about taking vitamin D supplements and HZ only included 126 patients with CKD, while the present study included more than 170 000 healthy participants.13 Compared with this previous study, the population of the study presented here was, largely, immunocompetent.
Another small, single hospital-based study in Taiwan compared people with post-herpetic neuralgia with matched hospital controls to assess the association between vitamin D and post-herpetic neuralgia. This cross-sectional analysis showed strong evidence that people with hypovitaminosis (defined as serum vitamin D <75 nmol/L) were associated with higher odds of having post-herpetic neuralgia (adjusted odds ratio = 3.12; 95% CI = 1.73 to 5.61).27 However, the cross-sectional analysis cannot distinguish temporality, which may lead to reverse causation. The authors’ cohort study has less potential for reverse causation, but it did not distinguish post-herpetic neuralgia from herpes zoster because of the limitation of the data.