Document resource
WHAT IS ALREADY KNOWN ON THIS TOPIC Large scale observational studies have found that cardiorespiratory fitness is strongly associated with type 2 diabetes in middle age, but whether fitness in early life is important for long term risk of type 2 diabetes is unclearA limitation of previous studies is their conventional analytical approach, largely susceptible to confounding, particularly from unobserved familial factors, limiting their ability to infer causalityWHAT THIS STUDY ADDS Higher levels of adolescent cardiorespiratory fitness were associated with a lower risk of type 2 diabetes in late adulthoodClinically relevant associations started from low levels of fitness, but with a smaller association in those with overweightComparing full siblings with different fitness levels, and hence controlling for all unobserved confounders that they share, the magnitude of the association was reducedHOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE, OR POLICY Adolescent cardiorespiratory fitness could be important in the development of type 2 diabetes in late adulthood, which might support and inform preventive interventionsRelying on conventional observational studies for answering causal questions about the role of adolescent cardiorespiratory fitness for later risk of type 2 diabetes might give biased estimates of the magnitude of the effectIntroduction Type 2 diabetes is a major public health concern, affecting at least half a billion people in 2019. 1 Together with the rising prevalence of overweight and obesity, a key contributor to the burden of type 2 diabetes,2 3 other modifiable risk factors, such as physical activity and its closely related trait, cardiorespiratory fitness, are trending downwards,4 5 particularly in younger age groups,6 7 and have worsened since the recent covid-19 pandemic.8–10 Because these traits track from a young age into adulthood,11 12 adolescence might represent an underused age group for preventive efforts.The current observational evidence linking cardiorespiratory fitness with type 2 diabetes is mainly based on studies in middle aged individuals,13 and much less is known about the role of fitness in adolescence, including whether it differs between individuals with overweight or normal weight.14 Also, previous studies were largely susceptible to confounding, particularly from unobserved familial factors, limiting their ability to infer causality. Because dealing with unobserved confounding in conventional observational analysis is challenging, triangulating the evidence with different methods is crucial to ensure reliable evidence that can support public health policy and interventions.15One method is family based analysis, where all shared factors, including unobserved behavioural, environmental, and genetic confounders, are inherently controlled by comparing family members (eg, siblings) who differ in terms of the exposure (eg, fitness) and outcome (eg, type 2 diabetes).16 To our knowledge, however, no previous study has used family based analysis, such as sibling comparisons, when examining the role of cardiorespiratory fitness in type 2 diabetes.We performed a nationwide cohort study of >1.1 million young men tested for cardiorespiratory fitness and who were prospectively followed for up to five decades for incidence of type 2 diabetes. To assess the validity of the observational findings, we performed sibling comparison analyses in half a million full siblings to partly deal with the currently unknown influence of unobserved confounding on the association.Methods Design This registry based cohort study was conducted by crosslinking data from nationwide data sources in Sweden, based on the personal identification number that is unique to all Swedish residents. The study was in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines. 17Data sources The study population was obtained from the Swedish Military Service Conscription Register 18 and included all men who participated in military conscription examinations from 1972 to 1995 and who had completed cardiorespiratory fitness testing. During this period, conscription at about age 18 years was mandatory for all Swedish men with few exemptions (ie, 90% of men were conscripted).18 Exemptions included specific functional disabilities, routine care needs, specific religious beliefs, or conscripting elsewhere. We used the Multi-Generation Register to identify all conscripts who were full siblings.19 From the National Patient Register, we collected data on diagnoses of type 2 diabetes from inpatient and specialist outpatient care.20 We collected data on dispensed antidiabetic drug treatments from the Prescribed Drug Register21 in an attempt to also identify primary care treatment or diagnosis of type 2 diabetes not covered by the patient register. Deaths were identified from the National Cause of Death Register.22 From databases managed by Statistics Sweden, we obtained data on emigration and socioeconomic data, and we also obtained information on biological sex, rather than from patient reported gender.19 23 Reporting to all these registers and databases is mandated by law.Derivation of study population Between 1972 and 1995, 1 249 131 Swedish men were conscripted. We excluded 33 645 (2.7%) men with missing cardiorespiratory fitness data, 72 042 (5.8%) with missing data on any covariates, and 19 395 (1.5%) with extreme values recorded for cardiorespiratory fitness and body mass index. Hence 1 124 049 men with complete data were included in the cohort analysis (90.0% retained, online supplemental figure 1). Among these, 477 453 were identified as full brothers from 219 304 families and were subsequently included in the full sibling analysis. Of these, 185 683 were two sibling families, 29 235 were three sibling families, and 4386 were families of four or more siblings.SP210.1136/bmjmed-2024-001313.supp2Supplementary dataAssessment of cardiorespiratory fitness The exposure was cardiorespiratory fitness, assessed during conscription with a maximal ergometer bicycle test and according to a standardised protocol. 18 The test results were recorded as watt maximum (Wmax), which was used as the main exposure in this study. Wmax has mainly been used in other studies based on Swedish conscriptin,24–28 and is the raw measure by which conscripts are compared during their military service. Briefly, conscripts performed the test after a normal electrocardiography procedure and began cycling for five minutes at 60-70 revolutions/min with a light external resistance, as determined according to body weight. The external resistance was gradually increased by 25 W/min until voluntary exhaustion. Wmax has been shown to have a strong correlation (r=0.88) with maximal oxygen uptake (ie, the gold standard metric of cardiorespiratory fitness in young people).29 We classified recorded values of <100 Wmax as extremely low and likely a result of data entry errors, and subsequently excluded these conscripts, as in previous studies.24 25 30Determining type 2 diabetes The outcome was type 2 diabetes, defined as a composite of a diagnosis of type 2 diabetes and dispensation of antidiabetic drug treatment, until 31 December 2023. The combination of diagnoses and drug treatments dispensed in establishing a diagnosis of type 2 diabetes has previously been used in Swedish registry studies. 31 Diagnoses were determined from 1997 onwards from the National Patient Register and the ICD-10 (international classification of diseases, 10th revision) code E11 in a primary or secondary position. An inpatient diagnosis of type 2 diabetes in the National Patient Register has a positive predictive value of 79-100%, although sensitivity is lower (23-84%).20 Information on dispensed antidiabetic drug treatments from 2005 onwards was obtained from the Prescribed Drug Register, based on the anatomical therapeutic chemical code A10.21Covariates Data on age at conscription, year of conscription, and body mass index were obtained from the Swedish Military Service Conscription Register. Body mass index was calculated based on height and weight, measured as part of the medical examination during conscription. We considered values <15 and >60 as extreme and likely a result of data entry errors, and subsequently excluded participants with these values, as in previous studies. 24 30 We collected socioeconomic data for both mothers and fathers of the conscripts from databases managed by Statistics Sweden.23 For education, we used the highest level attained during the individual's lifetime. For income, we extracted annual disposable income for each year between ages 40 and 50 years. Within each parental birth year cohort, we then calculated income quintiles at each age and identified the highest observed quintile across this age range. This approach gave a categorical variable reflecting the parent's peak working life income group, based on quintile. When values for both parents were available, we retained the highest value.Statistical analysis Participants were followed up from their conscription date to the date of a diagnosis of type 2 diabetes or dispensation of antidiabetic drug treatment, death, emigration, or end of follow-up (31 December 2023), whichever came first. Flexible parametric survival models were used to calculate hazard ratios for type 2 diabetes during follow-up by levels of fitness, with baseline knots placed at the 5th, 27.5th, 50th, 72.5th, and 95th centiles of the uncensored log survival times, and with age as the underlying timescale. 32 33 We modelled cardiorespiratory fitness as deciles (hereafter referred to as groups, with group 1 being the lowest fitness level and group 10 the highest) and with restricted cubic splines with knots placed at the 5th, 35th, 65th, and 95th centiles.33 To enhance the clinical interpretation and use of the results, we also computed the standardised cumulative incidence of type 2 diabetes at age 65 years (1−survival) and illustrated this incidence graphically over the follow-up period.32 Moreover, we calculated the preventable fraction of type 2 diabetes at age 65 years associated with different hypothetical public health interventions: a minor (moving all participants in fitness group 1 to fitness group 2), moderate (moving all participants in the bottom four fitness groups to fitness group 5), or extreme (moving all participants to fitness group 10) intervention.34 All models were adjusted for age at conscription (continuous), year of conscription (1972, 1973-77, 1978-82, 1983-87, 1988-92, and 1993-95), body mass index (continuous and quadratic term), parental education (compulsory school <9 years, secondary education, post-secondary education <3 years, and post-secondary education >3 years), and parental income (five categories). To explore whether body mass index acted as an effect modifier, we incorporated product terms between overweight status (body mass index ≥25.0 or <25.0) and the 10 groups of cardiorespiratory fitness into the model. We then computed marginal hazard ratios across overweight status for the total population while allowing for effect modification. Next, we post-estimated hazard ratios within strata of overweight status.For the full sibling analysis, the flexible parametric model was extended to a marginalised between-within model with robust standard errors (to account for the clustered nature of the data).16 35 The between-within model allows control of all unobserved confounders shared between siblings, including behavioural, environmental, and about 50% of genetic factors, by including an individual term for the exposure and covariates and a term for their family averages, hence separating the individual level variation (within effect) from the family level variation (between effect).16 35 Specifically, we included family average terms for cardiorespiratory fitness and for the observed covariates body mass index, age at conscription, and year of conscription. Family terms were not computed for parent level covariates because these are perfectly shared between siblings. Throughout the reporting of results from the full sibling analysis, hazard ratios represent the within effects.Sensitivity analyses A series of sensitivity analyses were conducted to examine the robustness of the results. We repeated the main analysis after additional adjustment for muscular strength, 13 and after adjustment for height,36 both of which have been implicated in type 2 diabetes and could influence performance on a cycle ergometer test. We then repeated the analysis after scaling Wmax to body weight (Wmax/kg), and after estimating maximal oxygen consumption (VO2max) from Wmax and body weight with a validated equation.37 To explore the potential for residual confounding within the categorical covariate conscription year, we repeated the analysis after modelling conscription year with restricted cubic splines with knots placed at the 5th, 35th, 65th, and 95th centiles. To determine whether changes in estimates from the cohort to the full sibling analysis were more likely because of selection bias in the sibling cohort rather than control for unobserved confounders,16 the standard cohort analysis was performed in the sibling cohort (ie, without controlling for shared factors). To assess whether the observed reduction in the magnitude of association from the cohort to the full sibling analysis was the same in twins, who are assumed to share more environmental and genetic factors, we repeated the analysis in all conscripts who were twins, defined as full siblings born in the same year and month, and contrasted these estimates with those derived from all non-twin full siblings (ie, the full sibling cohort excluding all twins). We also repeated the full sibling analysis by quintiles (five groups) of age differences between siblings in two sibling families.Because drug treatments for type 2 diabetes can also be used for other conditions (eg, polycystic ovary syndrome, obesity, heart failure, and type 1 diabetes), and information about the indication is not available in the Prescribed Drug Register,21 we also repeated the main analysis but only considered a patient registry recorded diagnosis of type 2 diabetes as the outcome. Also, because ICD-10 was not implemented officially until 1997, conscripts from earlier cohorts could have been followed for many years before receiving a diagnosis of type 2 diabetes, so we repeated the analyses restricted to those conscripted in 1985 or later. Finally, to relax the proportional hazards assumption, we also computed the standardised cumulative incidence of type 2 diabetes at ages 40, 45, 50, 55, 60, and 65 years, after allowing the effect of fitness to vary across time, using an interaction between a restricted cubic spline with three degrees of freedom of the follow-up time (33rd and 67th centiles of the distribution of the uncensored log survival times) and the 10 fitness groups. These estimates were compared with those derived from the main model (assuming proportional hazards). All analyses were performed with Stata MP version 16.1.Patient and public involvement Given the use of pseudonymised registry data, no patients or members of the public were involved in this study, and direct dissemination is not possible. However, findings will be communicated to the public through social media and may be subject to press releases through academic channels.Results Baseline characteristics In the total cohort, mean age at conscription was 18.3 (standard deviation 0.7) years, and most (81.3%) participants had a body mass index in the normal weight range ( table 1). About a quarter (26.6%) of participants had parents with a high (post-secondary) level of education and about a third (34.1%) had parents with a high (top category) level of annual income. Compared with those with the lowest fitness levels, those with higher fitness levels were, on average, born later, had a higher body mass index, and had parents with a higher level of education and income (online supplemental table 1). In the full sibling cohort, baseline characteristics were similar (table 1). The median age difference between full siblings in two sibling families was 3.4 (interquartile range (IQR) 2.3-5.2) years.Table 1Baseline characteristics in total and full sibling cohorts Total cohort(n=1 124 049)Full sibling cohort(n=477 453)Median (IQR) birth year1966 (1960 to 1971)1965 (1961 to 1970)Mean (SD) age at conscription (years)18.3 (0.7)18.3 (0.7)Mean (SD) body mass index21.8 (2.9)21.7 (2.8)Body mass index categories: Underweight (<18.5)88 966 (7.9)37 884 (7.9) Normal weight (18.5-24.9)913 586 (81.3)390 811 (81.9) Overweight (25.0-29.9)102 267 (9.1)41 248 (8.6) Obesity (>30.0)19 230 (1.7)7510 (1.6)Mean (SD) cardiorespiratory fitness (watt max)277 (52)277 (52)Parental level of education: Compulsory school <9 years319 242 (28.4)139 246 (29.1) Secondary education505 510 (45.0)210 246 (44.0) Post-secondary education <3 years123 757 (11.0)50 210 (10.5) Post-secondary education >3 years175 540 (15.6)77 751 (16.3)Parental highest income: Category 1 (low income)55 055 (4.9)18 918 (4.0) Category 2108 836 (9.7)41 820 (8.8) Category 3237 953 (21.2)101 150 (21.2) Category 4338 444 (30.1)145 735 (30.5) Category 5 (high income)383 761 (34.1)169 830 (35.6)Data are number (%) unless indicated otherwise.IQR, interquartile range; SD, standard deviation.Type 2 diabetes during follow-up During follow-up, 115 958 (10.3%) participants in the total cohort and 48 089 (10.1%) of the full siblings had a diagnosis of type 2 diabetes or had been dispensed antidiabetic drug treatment ( online supplemental figure 2). In the total cohort, median age at the first event was 53.4 (IQR 47.6-59.3) years (minimum 21.0, maximum 73.0). In the full sibling cohort, median age at the first event was 53.5 (IQR 47.8-59.1) years (minimum 21.0, maximum 72.5). Most events (93.4%) occurred after age 40 years, and about two thirds (65.0%) occurred after age 50 years. Most (78.3%) of the censoring was administrative censoring (ie, end of follow-up, online supplemental table 2).Cardiorespiratory fitness and type 2 diabetes In the cohort analysis after accounting for covariates, higher levels of cardiorespiratory fitness were linearly associated with a progressively lower risk of type 2 diabetes, starting from fitness group 2 ( figures 1–3, table 2, and online supplemental table 3). Compared with fitness group 1, the hazard ratio in fitness group 2 was 0.83 (95% confidence interval (CI) 0.81 to 0.85), with a difference in the standardised cumulative incidence at age 65 years of 4.3 (3.8 to 4.8) percentage points. The risk then gradually decreased, and the hazard ratio in fitness group 10 compared with fitness group 1 was 0.38 (95% CI 0.36 to 0.39), with an incidence difference of 17.8 (17.3 to 18.3) percentage points. A minor intervention, hypothetically moving all participants in fitness group 1 to fitness group 2, was estimated to prevent 7.2% (95% CI 6.4% to 8.0%) of type 2 diabetes events at age 65 years, whereas an extreme intervention, hypothetically moving all participants to fitness group 10, was estimated to prevent 35.6% (34.1% to 37.0%) of events (online supplemental table 4).Figure 1Hazard ratio for type 2 diabetes across restricted cubic splines of cardiorespiratory fitness in the cohort and full sibling analyses. Estimates were obtained with flexible parametric survival models, extended to a marginalised between-within model in the full sibling cohort, with baseline knots placed at the 5th, 27.5th, 50th, 72.5th, and 95th centiles of the uncensored log survival times, and with age as the underlying timescale. The knots for the splines of fitness were placed at the 5th, 35th, 65th, and 95th centiles. The referent was set to the median value of fitness group 1 (202 watt max). The models were adjusted for age at conscription, year of conscription, body mass index, parental education, and parental income. For graphical purposes, the x axis was limited to span from the first to the 99th centile of the exposure distribution. CI=confidence intervalFigure 2Standardised cumulative incidence of type 2 diabetes for the whole follow-up period by levels of cardiorespiratory fitness in the cohort and full sibling analyses. Cardiorespiratory fitness was categorised into deciles (referred to as groups, with group 1 being the lowest fitness level and group 10 the highest). Estimates were obtained with flexible parametric survival models, extended to a marginalised between-within model in the full sibling cohort, with baseline knots placed at the 5th, 27.5th, 50th, 72.5th, and 95th centiles of the uncensored log survival times, and with age as the underlying timescale. All models were adjusted for age at conscription, year of conscription, body mass index, parental education, and parental income. Confidence intervals were omitted for clarity because they are shown in figure 3, as well as in figure 1 and table 2Figure 3Difference in standardised cumulative incidence of type 2 diabetes at age 65 years by levels of cardiorespiratory fitness in the cohort and full sibling analyses. Cardiorespiratory fitness was categorised into deciles (referred to groups, with group 1 being the lowest fitness level and group 10 the highest). Estimates were obtained with flexible parametric survival models, extended to a marginalised between-within model in the full sibling cohort, with baseline knots placed at the 5th, 27.5th, 50th, 72.5th, and 95th centiles of the uncensored log survival times, and with age as the underlying timescale. Fitness group 1 (lowest fitness) was the referent. Values in parentheses show the median fitness in each group. All models were adjusted for age at conscription, year of conscription, body mass index, parental education, and parental incomeTable 2Hazard ratios for type 2 diabetes and differences in standardised cumulative incidence at age 65 years by levels of cardiorespiratory fitness in cohort and full sibling analyses. Cardiorespiratory fitness was categorised into deciles (referred to as groups, with group 1 being the lowest fitness level and group 10 the highest).Fitness groupCardiorespiratory fitness (Wmax) (median (range))Cohort analysis (n=1 124 049)Full sibling analysis (n=477 453)No of events/Total No of participants‡Hazard ratio (95% CI)Difference in standardised cumulative incidence at age 65 (percentage points, 95% CI)No of events/Total No of participants‡Hazard ratio (95% CI)Difference in standardised cumulative incidence at age 65 (percentage points, 95% CI)1202 (100-211)*18 302/120 001ReferenceReference7361/49 673ReferenceReference2222 (212-229)†15 508/105 3790.83 (0.81 to 0.85)−4.3 (−4.8 to −3.8)6426/44 7080.89 (0.85 to 0.94)−2.3 (−3.3 to −1.3)3240 (230-243)17 565/125 8920.76 (0.74 to 0.78)−6.1 (−6.6 to −5.7)7317/53 6100.85 (0.81 to 0.89)−3.3 (−4.2 to −2.3)4251 (244-257)12 111/101 0280.70 (0.68 to 0.72)−7.8 (−8.3 to −7.3)5036/43 2690.81 (0.77 to 0.86)−4.1 (−5.1 to −3.0)5264 (258-271)12 398/116 0910.64 (0.62 to 0.65)−9.7 (−10.2 to −9.2)5230/49 3130.75 (0.71 to 0.79)−5.6 (−6.6 to −4.5)6279 (272-286)10 131/106 2830.59 (0.58 to 0.61)−11.0 (−11.5 to −10.5)4343/45 8810.71 (0.67 to 0.75)−6.5 (−7.6 to −5.4)7294 (287-302)9479/113 7940.55 (0.54 to 0.57)−12.2 (−12.7 to −11.6)3939/48 1970.66 (0.62 to 0.70)−7.7 (−8.8 to −6.6)8313 (303-323)8205/111 8400.49 (0.48 to 0.50)−14.1 (−14.6 to −13.6)3400/48 0710.60 (0.57 to 0.64)−9.1 (−10.3 to −8.0)9333 (324-345)6332/111 8890.44 (0.42 to 0.45)−15.8 (−16.4 to −15.3)2617/46 9830.58 (0.54 to 0.62)−9.7 (−10.9 to −8.5)10367 (346-999)5927/111 8520.38 (0.36 to 0.39)−17.8 (−18.3 to −17.3)2420/47 7570.53 (0.50 to 0.57)−10.9 (−12.1 to −9.7)Models were adjusted for age at conscription, year of conscription, body mass index, parental education, and parental income.*Median in fitness group 1 in the sibling cohort was 203.†Median in fitness group 2 in the sibling cohort was 223.‡First type 2 diabetes event.CI, confidence interval.In the full sibling analysis, the association between cardiorespiratory fitness and risk of type 2 diabetes was replicated, although with a reduction in the magnitude of the association (figures 1–3, table 2, and online supplemental table 3). Compared with fitness group 1, the hazard ratio in fitness group 2 was 0.89 (95% CI 0.85 to 0.94), with an incidence difference of 2.3 (1.3 to 3.3) percentage points. In fitness group 10, the hazard ratio was 0.53 (0.50 to 0.57), with an incidence difference of 10.9 (9.7 to 12.1) percentage points compared with fitness group 1. A minor intervention was estimated to prevent 4.6% (2.6% to 6.5%) of type 2 diabetes events at age 65 years, whereas an extreme intervention was estimated to prevent 24.3% (20.5% to 28.0%) of events (online supplemental table 4). Allowing for interaction effects across overweight status suggested possible effect modification, with a smaller association in individuals with overweight than in those without overweight, particularly in the full sibling analysis (figure 4 and online supplemental table 5).Figure 4Post-estimated hazard ratio for type 2 diabetes across restricted cubic splines of cardiorespiratory fitness within strata of overweight status in the cohort and full sibling analyses. Estimates were obtained with flexible parametric survival models, extended to a marginalised between-within model in the full sibling cohort, in the total sample, and subsequently post-estimated within strata of overweight status using standardisation (G formula). Baseline knots were placed at the 5th, 27.5th, 50th, 72.5th, and 95th centiles of the uncensored log survival times, and with age as the underlying timescale. The knots for the splines of fitness were placed at the 5th, 35th, 65th, and 95th centiles. The referent was set to the median value of fitness group 1 (202 Wmax). The models were adjusted for age at conscription, year of conscription, parental education, parental income, overweight status, and product terms between overweight status and fitness groups. For graphical purposes, the x axis was limited to span from the first to the 99th centile of the exposure distribution. CI=confidence intervalSensitivity analyses The results were not substantially influenced by adjustment for muscular strength ( online supplemental table 6 and 7), height (online supplemental table 8), when rescaling the exposure (online supplemental table 9), or when modelling year of conscription using restricted cubic splines (online supplemental table 10). Applying the standard analysis in the full sibling cohort produced similar estimates as in the cohort analysis (online supplemental table 11). Overall, the estimates derived from the twin analysis were similar to estimates from the non-twin full sibling analysis (online supplemental table 12), although the analysis was characterised by limited precision because of lower statistical power in this analysis. Furthermore, the estimates were generally similar when examined in five groups of within family age differences (online supplemental table 13). Based on an outcome definition that only included patient registry recorded diagnoses of type 2 diabetes (online supplemental table 14 and supplemental figure 3), and restricting the analysis to those who were conscripted in 1985 or later (online supplemental table 15), gave overall similar results as in the main analysis. Finally, allowing the effect of fitness to vary over time produced similar estimates as when assuming proportional hazards from age 50 to 65 years, but the incidence differences were smaller at age 40 and 45 years (online supplemental table 16).Discussion Principal findings In this nationwide cohort study spanning six decades, higher levels of cardiorespiratory fitness in late adolescence were associated with a progressively lower risk of type 2 diabetes in late adulthood, with clinically relevant associations starting from low levels of fitness. The association was replicated after adjusting for unobserved behavioural, environmental, and genetic confounders shared between full siblings, but the magnitude of association was reduced, particularly in those with overweight.Comparing fitness group 2 with fitness group 1, we found a 4.3 percentage point difference in the standardised cumulative incidence of type 2 diabetes at age 65 years, corresponding to a hypothetically preventable fraction of about 7%. We found little evidence that the association levelled off at higher fitness levels. Comparing fitness group 10 with fitness group 1, we found an incidence difference of 17.8 percentage points, and hypothetically moving all participants to fitness group 10 was estimated to prevent about a third of type 2 diabetes events. Estimating preventable fractions associated with hypothetical interventions implies strong assumptions, however, and whether these effects can be realised from public health interventions is uncertain. For example, although controlled interventions for regular physical activity (exercise) can cause improvements in fitness of a magnitude comparable with the difference in fitness between fitness groups 1 and 2 in our study (a difference of about 10%),38 whether such effects can be achieved outside controlled settings is not clear. A less controlled example is school based physical activity interventions, which in meta-analyses have been shown to possibly have a small effect on VO2max in children and adolescents (mean difference 1.19 mL/kg/min, 95% CI 0.57 to 1.82; low certainty of evidence).39Comparison with other studies Although previous studies have typically measured cardiorespiratory fitness in middle age and with limited follow-up, 13 our study provides new evidence on the association across the life course of participants. Our findings suggest that adolescent cardiorespiratory fitness might be important in the development of type 2 diabetes, and particularly for long term risk. When we modelled the time dependent effect of fitness, the associations seemed to be more pronounced in late adulthood (ie, age ≥50 years) than in early adulthood. Our findings might support and inform preventive interventions delivered to specific target groups (ie, those with the lowest or declining fitness levels), which could be identified through various surveillance programmes.40 Our findings also build on those from a previous study based on the same population which also estimated strong associations between adolescent cardiorespiratory fitness and future type 2 diabetes.14 One of the strengths of our study is that we examined the previously unknown role of familial confounding, while also following a larger proportion of our study population into older ages when type 2 diabetes is typically diagnosed.41An important finding of this study was therefore that the association between levels of adolescent cardiorespiratory fitness and risk of type 2 diabetes in late adulthood was replicated after adjusting for unobserved familial confounders shared between full siblings, but with a reduction in magnitude. Comparable studies are few, and the findings are inconsistent. For example, in one study, twin participants who were physically active had a lower risk of type 2 diabetes than their inactive co-twin (hazard ratio 0.54, 95% CI 0.37 to 0.78).42 The results were similar in a small sample of monozygotic twins, although with a greater degree of uncertainty because of limited statistical power.42 Two studies used a mendelian randomisation framework, which in theory can be more robust to confounding than a conventional observational analysis. One of these studies did not link genetically predicted cardiorespiratory fitness with type 2 diabetes,43 although the study was a limited two sample analysis based on summary statistics. In contrast, in an individual participant mendelian randomisation analysis, focusing specifically on cardiorespiratory fitness and incidence of type 2 diabetes, and with more thorough assessment of the underlying assumptions, higher fitness levels were associated with a lower risk of type 2 diabetes in both conventional and mendelian randomisation analyses.44Notwithstanding these strengths, because the data came from the UK Biobank (ie, middle aged individuals), and mendelian randomisation estimates are typically interpreted as lifetime effects of the exposures, these findings do not necessarily provide evidence about the causal effects of adolescent fitness. Moreover, interpreting mendelian randomisation findings for public health is challenging considering the lack of measures that can more easily be contextualised, such as standardised incidences. In this respect, our study adds important evidence because the difference in magnitude of association between the cohort and full sibling analyses seemed to be clinically relevant. For example, the differences in incidence between fitness groups were about 40% smaller in sibling comparison analyses than in the cohort analysis, and the share of preventable events associated with hypothetical interventions of different intensities was reduced by about a third. We might assume that this finding to some degree reflects control for shared familial confounders. Our sibling analysis cannot determine to what extent the observed reduction is attributable to control for genetic or shared environmental confounders, although our sensitivity analysis provided some indications that genetics might be the driving force. This result might be supported by genetically informed studies, such as those based on linkage disequilibrium score regression, showing negative genetic correlations between cardiorespiratory fitness and risk factors for type 2 diabetes,45 and between physical activity and type 2 diabetes,46 which would bias traditional observational evidence in favour of fitness.Before definitive conclusions can be drawn, however, future studies with higher statistical power, including studies with data on twin zygosity, are needed. Considered alongside the results from other family based and genetically informed studies, our findings are consistent with adolescent cardiorespiratory fitness as a potential risk factor for future type 2 diabetes. At the same time, conventional observational studies might give biased estimates of the magnitude of the effect because of negative familial confounding, possibly explained mainly by genetic factors.Study implications The implications of this study can be further contextualised by considering evidence from randomised trials, where the efficacy of combined lifestyle interventions against type 2 diabetes onset has been shown in individuals with an increased risk of type 2 diabetes. 47 But whether early life cardiorespiratory fitness has a preventive role in the risk of type 2 diabetes in later life remains unclear. Our findings indicate that adolescent cardiorespiratory fitness might be important in the development of type 2 diabetes in late adulthood, but with some evidence of possible effect modification by overweight status. Specifically, the association seemed to be smaller in those with overweight than in those without overweight. This finding is important, given the rising prevalence of overweight and obesity, including in young people, which is a key contributor to the burden of type 2 diabetes.1 3 48Limitations and strengths of this study Our study had several limitations. Firstly, although the Swedish Military Service Conscription Register offers unique advantages for studying the role of adolescent fitness in future health outcomes, because of its standardised data collection and broad coverage, historical conscription practices limited participation to men. Hence our findings might not be applicable to women. Some evidence exists that the association between cardiorespiratory fitness and type 2 diabetes could be slightly stronger in women than in men. 13 Further causal analyses on cardiorespiratory fitness and type 2 diabetes in women, including those of family based designs, are warranted.Secondly, although we controlled for observed confounders, such as body mass index and socioeconomic factors, as well as unobserved familial confounders through sibling analysis, residual and unobserved confounding could remain (eg, from non-shared genes and behaviours). Thirdly, the sibling comparison design also relies on assumptions, where estimates that reduce towards the null might be caused by amplified bias from non-shared confounding, measurement error, or control for shared mediators.16 49 Examples of potential non-shared confounders could be different factors during childhood and pre-adolescence that might have caused one sibling to have a lower fitness level and higher risk of future type 2 diabetes than their sibling, such as health problems early in life (eg, infections). Moreover, the assumption of non-shared confounding should be considered, and its subsequent risks, when interpreting the findings from the sensitivity analysis of siblings with larger age differences; non-shared confounding is presumably greater in these sibling pairs. Future studies based on other types of causal analysis would be valuable.Fourthly, we identified diagnoses of type 2 diabetes after ICD-10 was implemented (in 1997), meaning that some degree of overestimation of age of first diagnosis was likely in our study because patients treated earlier might not have been captured until the data became available to us. The results were similar in analyses restricted to participants from the later conscription cohorts, however, and this left truncation of the type 2 diabetes data is unlikely to have influenced the different estimates in the cohort versus sibling analyses. Nonetheless, the incidence might be underestimated if conscripts received a diagnosis before 1997 or were dispensed drug treatments before 2005 and went into remission later and none of their subsequent records noted a history of diabetes. For most individuals, however, we believe that this possibility is unlikely.Fifthly, because the National Patient Register does not cover primary care, a risk of differential misclassification bias exists if those who received a diagnosis in primary care (and not included in our study) had a better health status (eg, greater fitness) or a less severe form of type 2 diabetes, or both, potentially overestimating the association. We attempted to capture some primary care diagnoses by using nationwide data on dispensed antidiabetic drug treatments, although the Prescribed Drug Register was not implemented until 2005. Furthermore, reliance on routinely collected data, particularly the use of antidiabetic drug treatments as an indication, rather than clinical verification, might introduce some misclassification, including some diagnoses of secondary diabetes or adult onset type 1 diabetes, although these diagnoses are likely rare compared with primary type 2 diabetes. Reassuringly, however, the associations were similar for the composite outcome of diagnosis of type 2 diabetes and dispensed antidiabetic drug treatments versus diagnosis of type 2 diabetes only. Finally, our analyses were based on historical data from one country, and the extent to which these estimates might differ across countries or vary with secular trends in fitness levels, cardiometabolic risk factors, and risk management practices warrants further investigation.Our study had several strengths, including the large sample of young men with standardised measures of cardiorespiratory fitness and prospective follow-up over several decades based on nationwide registers with practically no attrition. Moreover, the study had substantial statistical power which allowed us to perform a comprehensive set of analyses, including sibling comparisons in a large sample of full siblings.Conclusions The results of our study suggest that higher levels of cardiorespiratory fitness in late adolescence are associated with a progressively lower risk of type 2 diabetes in late adulthood, with clinically relevant associations starting at low levels of fitness. The association was replicated after adjustment for unobserved familial confounders shared between full siblings, but the magnitude of the association was reduced, particularly in those with overweight. These findings suggest that adolescent cardiorespiratory fitness could be important in the development of type 2 diabetes in late adulthood, but that conventional observational analysis might give biased estimates of the magnitude of the effect.SP110.1136/bmjmed-2024-001313.supp1Supplementary data