Author's reply to “Parallel trends are not causal evidence: methodological caution regarding the pediatric obesity–youth-onset type 2 diabetes link”
Article information
To the editor,
We thank the correspondent for a careful and constructive reading of our review and for the collegial tone of the critique. We agree with the central clinical message the letter affirms—that primary prevention of childhood obesity remains the most powerful available strategy against youth-onset type 2 diabetes mellitus (T2DM)—and we welcome the opportunity to clarify several points of calibration. As the review has already been published, we address the substantive and interpretive points below within this reply, and we separately confirm which items require a formal correction to the published record.
On causal language and the ecological design
We accept the methodological point that our own synthesis—a narrative, ecological comparison of populationlevel trajectories—cannot by itself establish direction of effect, and that Fig. 1 was an illustrative visual summary rather than a formal lag or cross-correlation test. We would clarify, however, that the causal interpretation offered in our discussion did not rest on the ecological comparison alone; it was intended to draw on the wider body of individual-level and quasi-experimental evidence that is not subject to the same confounding. Two lines of evidence are worth setting out explicitly here, as they were underemphasized in the original discussion. First, Mendelian randomization studies, which use germline genetic variants as instruments to avoid reverse causation and shared confounding, have consistently found that genetically elevated body mass index (BMI) increases T2DM risk: Geng et al. [1] reported that each 1-standard deviation (SD) increase in genetically instrumented childhood BMI was associated with an 83% increase in adult T2DM risk (odds ratio [OR], 1.83; 95% confidence interval [CI], 1.46–2.30), and independent Mendelian randomization (MR) analyses in adults have reported comparable causal odds ratios per unit BMI increase [2,3]. These genetic-instrument analyses are far less vulnerable to the reverse-causation and shared-confounder concerns the correspondent correctly raises about ecological time trends. Second, and more directly relevant to a causal reading, interventional evidence shows that deliberately reducing adiposity reverses the glycemic abnormality, which is the strongest available test of a causal relationship in humans. The Diabetes Remission Clinical Trial (DiRECT), a cluster-randomized trial of a primary care-delivered weight-management program, achieved diabetes remission in 46% of intervention participants at 12 months, with remission closely tracking the amount of weight lost (a weight loss of >15 kg predicted the great majority of remissions), and this dose-response relationship between weight loss and glycemic reversal was sustained at 24 months [4,5]. Bariatric surgery series (Swedish Obese Subjects, Surgical Treatment and Medications Potentially Eradicate Diabetes Efficiently [STAMPEDE]) and the pediatric Teen-Longitudinal Assessment of Bariatric Surgery (Teen-LABS) cohort we cited show the same pattern of reversal following surgical weight loss. We agree that this quasi-experimental and genetic evidence, rather than the visual parallelism of Fig. 1 alone, is the more defensible basis for causal language, and we will foreground it explicitly should the opportunity for revision arise (for example, in a future update, correspondence, or related work); within the published text itself, we would ask that this reply serve as the accompanying clarification, per usual practice for a “Correspondence and reply” pair.
On the Israel and Japan data points
We agree with this observation. Israel's 441% relative increase arises from a low baseline incidence (<1/100,000/yr) and is therefore highly sensitive to improving case ascertainment and registry completeness over the same interval, while Japan's data derive from a school-based urine glucose screening program that is methodologically distinct from registry-based ascertainment elsewhere. Both points are well taken, and we did not sufficiently flag either as a methodological outlier in the original discussion. We record this limitation here for the published record: readers should weigh the Israeli and Japanese data points with appropriate caution rather than treating them as directly comparable to the registry-based estimates from the USA, Canada, England and Wales, and Australia.
On "necessary and sufficient" and the Teen-LABS denominator
We agree this phrasing overstated our own findings. Approximately 25% of affected youth globally (and closer to 32% in Japan) are not obese by BMI criteria at diagnosis, and we ourselves noted that roughly 12% of apparent T2DM presentations may harbor monogenic (maturity-onset diabetes of the young [MODY]) diabetes; both figures are inconsistent with strict necessity or sufficiency. We clarify here that adiposity should be understood as the dominant, but not universal, phenotypic substrate for youth-onset T2DM. We also agree that a supporting citation for the 12% MODY-misclassification estimate was omitted from the original text; the relevant supporting literature includes Kleinberger and Pollin's review of monogenic diabetes misclassification in patients with an apparent type 2 phenotype [6]. We likewise agree that citing the Teen-LABS remission rate as 95% without its 19/20 evaluable-participant denominator, as we did in the abstract, understates the imprecision of that estimate; we record the correct denominator here (19 of 20 evaluable participants with T2DM achieved remission at 3 years) for the published record.
On the 2 citation errors
We thank the correspondent for identifying these and confirm both are genuine errors that arose during reference renumbering: the claim regarding upward trends in England and Wales was miscited to a Finnish type 1 diabetes cohort study rather than to the National Paediatric Diabetes Audit data summarized in Perng et al. [7], and the corresponding claim for New Zealand was miscited to an Australian type 1 diabetes clinical care guideline rather than to the same source. Because these are factual citation errors in the published article rather than matters of interpretation, we will separately request that the Editorial Office issue a formal erratum correcting both citations to reference Perng et al. [7] in Table 1 and the accompanying text, in line with the journal's correction policy.
On quality appraisal traceable to secondary sources
We agree with this observation and acknowledge it here. Primary data for the England and Wales, Israel, Australia, New Zealand, and Japan registries in Table 1 were drawn from the narrative review by Perng et al. [7] rather than independently reappraise from each primary report, so the Newcastle-Ottawa Scale scores of 7/9–8/9 assigned to those rows reflect the quality of that secondary synthesis rather than a de novo appraisal of the underlying registry publications. We record this as a limitation of the published review and, where the primary reports are accessible, intend to reappraise them directly and report the results in future correspondence or follow-up work.
In summary, we are grateful for a critique that sharpens rather than undermines the review's central conclusion. Of the points raised, the 2 citation errors are factual matters that we will pursue through a formal erratum with the Editorial Office. The remaining points are matters of interpretation and emphasis, which we have addressed directly in this reply: we have clarified that the causal language in our discussion was intended to rest on Mendelian randomization and interventional weight-loss evidence rather than on the ecological trend comparison alone, we have flagged Israel and Japan as methodologically distinct data points, we have softened the "necessary and sufficient" claim and supplied the omitted MODY citation and Teen-LABS denominator, and we have acknowledged that quality scores for 5 registries trace to a secondary synthesis. None of this alters the review's central public health message: pediatric obesity remains the dominant, modifiable upstream driver of youth-onset T2DM, and its primary prevention remains the single most powerful strategy available to arrest this epidemic.
Supportive evidence for the causal obesity–T2DM relationship
For the reader's convenience, we summarize here the principal longitudinal and interventional studies supporting a causal, rather than merely parallel, relationship between obesity and T2DM, spanning genetic, observational, and interventional designs:
• Mendelian randomization (genetic causal inference): Geng et al. [1] found that a 1-SD genetically instrumented increase in childhood BMI raised adult T2DM risk by 83% (OR, 1.83; 95% CI, 1.46–2.30), a design immune to reverse causation and most confounding. Corbin et al. and subsequent wide-angled MR analyses of adult BMI similarly confirmed a causal odds ratio for T2DM per unit BMI increase [2,3].
• Diet-induced weight loss reversing glycemia (randomized controlled trial [RCT] evidence): the DiRECT trial achieved diabetes remission in 46% of participants at 12 months and in over a third at 24 months through a structured primary care weight-management program, with remission probability rising in direct proportion to the amount of weight lost—a dose-response relationship that is difficult to explain other than causally [4,5].
• Bariatric surgery reversing glycemia (RCT and cohort evidence): the Swedish Obese Subjects study and the STAMPEDE randomized trial both demonstrated superior glycemic control and diabetes remission after surgical weight loss compared with medical therapy alone, and our own review cited the pediatric Teen-LABS cohort, in which 19 of 20 evaluable adolescents with T2DM (95%) achieved remission at 3 years after bariatric surgery [8].
• Pharmacological weight loss improving glycemia (RCT evidence): Glucagon-like peptide-1 receptor agonist trials (Semaglutide Treatment Effect in People with obesity [STEP] TEENS, liraglutide NN8022-4180) already cited in our review demonstrated that pharmacologically induced weight loss in adolescents produced concurrent improvements in hemoglobin A1c and fasting glucose, again consistent with a causal, reversible relationship between adiposity and dysglycemia rather than a coincidental parallel trend.
Taken together, this genetic and interventional evidence—rather than the ecological registry comparison alone—is, in our view, the more appropriate basis for causal language regarding the obesity–T2DM relationship, and we are grateful to the correspondent for prompting us to make this distinction explicit.
Notes
Conflicts of interest
No potential conflict of interest relevant to this article was reported.
Funding
This study received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.