Comments on Rahmati et al., The global impact of COVID‐19 pandemic on the incidence of pediatric new‐onset type 1 diabetes and ketoacidosis: A systematic review and meta‐analysis. J Med Virol. 2022; 1‐16 (doi: 10.1002/jmv.27996)

Abbreviations

COVID‐19

coronavirus disease 2019

HbA_1c

glycated hemoglobin

log

natural logarithm

logit(p)

log(p/(1−p))

T1D

type 1 diabetes

(s)DKA

(severe) diabetic ketoacidosis

AUTHOR CONTRIBUTIONS

Joachim Rosenbauer conceptualized and performed the data analyses and wrote the first draft of the manuscript. Anna Stahl wrote the first draft of the manuscript. Sabrina Schlesinger and Oliver Kuß critically supervised the analysis. All authors contributed to data interpretation and revision of manuscript drafts and read and approved the final manuscript. Joachim Rosenbauer is the guarantor of this work and, as such, takes responsibility for the integrity of the data and the accuracy of the data analysis.

CONFLICT OF INTEREST

The authors declare no conflict of interest.

Dear Editor,

We read with interest the meta‐analysis by Rahmati ¹ on the impact of the COVID‐19 pandemic on the incidence and presentation at onset of pediatric type 1 diabetes (T1D). The meta‐analysis compared pandemic data from 2020 to prepandemic data from 2019. However, the presentation of methods and results raises some questions. Furthermore, we noted some issues when trying to reproduce the results. We recalculated the meta‐analyses using Stata 17. ²

Rahmati ¹ insufficiently described the effect sizes used in the meta‐analysis for the number of cases and incidence rate of new‐onset T1D. Presumably, the event rate used in the effect size in figure 2A in Rahmati ¹ (logit(event rate) = natural logarithm (log) of (event rate)/(1 − event rate)) was the proportion of cases in 2020 related to the total number of cases in 2019 and 2020. Thus, the effect size actually equals log(cases in 2020/cases in 2019). The effect size estimates given in figure 2B in Rahmati ¹ are consistent with log(incidence rate in 2020/incidence rate in 2019), although confusingly termed logit(event rate). For both outcomes, the results would be easier to understand and interpret if the effect sizes were given on the nonlogarithmic scale (case ratio, incidence rate ratio). Last, differentiation between cohort and cross‐sectional studies seems unreasonable because the design of all studies is actually similar in comparing pandemic data with prepandemic data.

According to the methods, Rahmati ¹ used random‐effects models to estimate pooled effects. We attempted to reproduce the estimates presented in figures 2–4 in Rahmati ¹ using the estimates, standard errors, or confidence intervals of effect sizes given for the individual studies in these figures. The results in Table 1 suggest that pooled estimates regarding T1D cases, T1D incidence rate, diabetic ketoacidosis (DKA) prevalence, and glucose given in Rahmati ¹ are derived from fixed‐effects models. Of note, pooled effects for the number of new‐onset T1D cases and glucose level at onset were not statistically significant according to the 95% confidence interval (CI) from the DerSimonian‒Laird random effects model.

Table 1.

Meta‐analyses of effect estimates regarding outcomes in the pandemic (2020) versus the pre‐pandemic (2019) period—Reanalysis of studies included in Rahmati 2022 ¹

Note: For each outcome, the studies included are those included in Rahmati 2022, ¹ if not stated otherwise. Gray shading indicates that respective estimates match Rahmati's estimates. (s) DKA: (severe) diabetic ketoacidois.

^aDerSimonian‒Laird model.

^bEvent rate equals proportion of cases in 2020 related to the total number of cases in 2019 and 2020.

^cRatio of number of cases in 2020 versus 2019.

^dEstimated as exponential function of logit(event rate) reported in Rahmati 2022. ¹

^eLogarithm of the ratio of incidence rate in 2020 versus 2019 (termed logit(event rate) in fig. 2A in Rahmati 2022 ¹ ).

^fRatio of incidence rate in 2020 versus 2019.

^gRatio of (s)DKA prevalence in 2020 versus 2019 (termed risk ratio in fig. 3 in Rahmati 2022 ¹ ).

^hStudy of Lee 2022 ³  excluded.

ⁱData corrections for the studies Hernandez Herrero 2021, ⁴  Salmi 2022, ⁵ and Dilek 2021. ⁶

^jStudy of Bogale 2021 ⁷  additionally included (not considered in Rahmati 2022 ¹  due to a putative missing value).

Furthermore, Rahmati ¹  reported in figure 3 risk ratios (prevalence ratios) for (severe) DKA from individual studies that seriously deviate from original estimates for almost all studies. We re‐evaluated the prevalence of (severe) DKA using the number of (severe) DKA cases at onset in the pandemic and the prepandemic periods given in Table 1 of Rahmati ¹ for the individual studies. However, we revised some figures to resemble the data given in the original publications. Supporting Information: Table S1 gives details of the data correction. Corrected data and study‐specific and pooled (DerSimonian‒Laird random‐effects model) prevalence ratios with 95% CI are shown in Figure 1. The overall DKA prevalence was 37.0% (95% CI: 22.9%−52.7%) higher during the pandemic than during the prepandemic period, while Rahmati ¹ reported a distinctly lower increase of 10.2% (6.9%–13.5%). The overall severe DKA prevalence was 66.5% (45.9%−90.1%) higher in the pandemic period than in the prepandemic period, while Rahmati ¹ reported a distinctly different increase of 5.5% (3.3%−7.9%). When the study of Bogale ⁷ (not considered by Rahmati due to a putative missing value) was included, the prevalence increase of DKA (35.2% [21.5%−50.4%]) and severe DKA (62.4.0% [40.1%−88.3%]) was slightly attenuated (Supporting Information: Figure S1).

Figure 1.

Random‐effects meta‐analysis of the risk of (A) diabetic ketoacidosis (DKA) and (B) severe diabetic ketoacidosis (sDKA) at onset of pediatric type 1 diabetes in the pandemic versus the pre‐pandemic period—Reanalysis of studies included in fig. 3A of Rahmati 2022. ¹ Some numbers of (s)DKA/T1D cases used in Rahmati 2022 ¹ were revised to resemble the data given in the following original publications: Boboc 2021, ⁸ Goldman 2022, ⁹ Gottesmann 2022, ¹⁰ Hawkes 2021, ¹¹ Hernandez Herrero 2021, ⁴ Ho 2021. ¹² Supporting Information: Table S1 gives details of the data correction. Prevalence ratio: Ratio of DKA prevalence in pandemic versus pre‐pandemic period (termed risk ratio in Rahmati ¹ ). τ ²: between‐study variance. I ²: statistic of between‐study heterogeneity. H ²: statistic of between‐study heterogeneity (=1/(1 − I ²)). Q, p: Cochran's Q statistic for testing heterogeneity between studies and corresponding p value. z, p: z‐statistic for testing the hypothesis of a zero pooled effect and corresponding p value. Q_b, p: test statistic for testing between‐group differences of the pooled effect and corresponding p value.

Moreover, Rahmati ¹ used questionable effect sizes regarding the T1D incidence rate for three studies. ⁴ , ⁵ , ⁶ The effect size used for a Finnish study relates to the incidence rate of pediatric intensive care unit admission and not to the T1D incidence rate (Table 1). ⁵ Again, the effect size used for a Turkish study is not consistent with the original incidence rate data. ⁶ Finally, from a Spanish study, Rahmati ¹ selected incidence rate data that did not refer to the total population but to a selective population with a Northwest African (Maghreb) immigrant background (Table 1). ⁴ When the corrected data for these studies are used in a random‐effects model, the incidence rate during the pandemic was 13.5% (3.7%−24.3%) higher than that in the prepandemic period and not 63.7% (33.5%−100.8%) higher, as reported by Rahmati ¹ (Table 1).

Finally, Rahmati ¹ included the study of Lee ³ in the meta‐analysis regarding new‐onset T1D cases (figure 2A) that provides no information on the change in new‐onset cases during the pandemic, because it included age‐matched new‐onset T1D cases from the pandemic and the prepandemic periods. Excluding the study from the meta‐analysis affected the pooled estimate only slightly due to the small inverse‐variance weight of this study (Table 1).

Beyond the limitations mentioned by Rahmati, ¹ a further limitation of their meta‐analysis is that the certainty of evidence was not assessed according to the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach. ¹³ Furthermore, significant heterogeneity was found between studies possibly due to background risk population, different time periods studied, and differing COVID‐19 waves and respective containment measures. Although most studies had high quality according to the Newcastle–Ottawa scale relevant sources of bias cannot be excluded. In addition, pooled effect estimates for some outcomes are based on few studies only limiting the evidence. Due to the limitations, results should be interpreted cautiously.

In conclusion, the meta‐analysis of Rahmati ¹ significantly overestimated the increase in the pediatric T1D incidence rate but distinctly underestimated the prevalence increase of (severe) DKA at pediatric T1D onset.

Supporting information

Supplementary information.

Supplementary information.

Supplementary information.

DATA AVAILABILITY STATEMENT

The data used are available from the meta‐analysis of Rahmati et al. (2022) or the original publications included in this meta‐analysis.