The stakes reach far beyond the exercise laboratory. Type 1 diabetes arises when the immune system destroys the insulin-producing beta cells of the pancreas, committing patients to lifelong insulin therapy. Even under good metabolic control, cardiovascular disease remains a leading cause of death in this population, and the global prevalence and societal burden of the disease are climbing rapidly. Yet much of the prevention guidance these patients receive is extrapolated from studies of type 2 diabetes, a condition with fundamentally different machinery — insulin resistance rather than autoimmune beta-cell loss — and the Finnish team argues that borrowed evidence of this kind may not transfer cleanly. Their question was therefore sharply defined: strip away the neuropathy, nephropathy, retinopathy, hypertension, smoking, cardiovascular medication, and major comorbidities that confound most diabetes cohorts, and does uncomplicated type 1 diabetes itself leave any measurable imprint on how the heart, lungs, and circulation perform under maximal physiological stress?

Earlier research gave that question real teeth. Cardiopulmonary exercise testing, or CPET, is physiology’s most integrated stress probe: with a subject on a cycle ergometer, it captures breath-by-breath gas exchange, ventilatory patterns, and — in this study — noninvasive cardiac function as workload climbs. Previous studies had reported lower maximal oxygen uptake in type 1 diabetes, blunted oxygen pulse, reduced tidal volume and expiratory flow at peak exercise, and a heart-rate curve that rises almost linearly with workload instead of the normal inverted pattern in which heart rate surges at low intensity and then flattens toward maximum. That linear signature has been interpreted as a possible fingerprint of diabetes-related myocardial change, and it has appeared even in patients without long-term complications. But the literature is riddled with inconsistencies: some cohorts show lower oxygen uptake, others show none, and glycated hemoglobin alone cannot explain the scatter of results. Most tellingly, earlier work rarely tracked stroke volume — the volume of blood ejected with each heartbeat and the true engine of cardiac output — leaving the central hemodynamic story unresolved.

The answer came from the DIAMES project — Effects of Exercise and Stress on Glucose Metabolism in Type 1 Diabetes — conducted between 2021 and 2022 under approval from the ethics committee of the Northern Savo Hospital District. Fifty-three volunteers began; after excluding three participants with diabetes who used hypertension medication and two controls whose electrocardiograms were too corrupted for reliable heart-rate analysis, the final cohort comprised 15 adults with uncomplicated type 1 diabetes and 33 controls matched for age, sex, height, and body mass index. The screening was deliberately ruthless: no neuropathy, nephropathy, or retinopathy; no asthma, hypertension, or coronary disease; no renal insufficiency, pacemaker, or cardiovascular medication; no smoking. The participants with diabetes had lived with the condition for an average of fifteen years and maintained an average HbA1c of 57.9 mmol/mol — roughly 7.5 percent — with half managed by insulin pump and half by multiple daily injections. All were physically active, and total weekly exercise did not differ between the groups.

On test days, participants arrived at the university’s HUMEA laboratory between seven in the morning and two in the afternoon, having abstained from alcohol, caffeine, and tobacco for twelve hours and from intense training for twenty-four. Blood glucose was confirmed to sit between 5 and 13.9 mmol/L before the protocol and was monitored by finger-prick throughout, in line with international standards for exercise in diabetes. The test itself was an incremental cycling CPET: three minutes of seated rest, then three-minute stages of increasing power, rising by 35 watts per stage for men and 25 watts for women — sex-specific scaling designed to yield a fatigue-limited finish in roughly eight to twelve minutes, so that cardiopulmonary limits rather than exhausted quadriceps would decide the endpoint. Expired gases were measured breath by breath with a calibrated portable spiroergometer while a two-lead electrocardiogram sampled at 2,000 hertz tracked every beat. The methodological centerpiece was impedance cardiography, which estimates stroke volume noninvasively from beat-to-beat changes in thoracic electrical impedance as each contraction ejects blood — a technique validated against the direct Fick method at rest and during dynamic exercise. Peak effort was verified by heart rates beyond 95 percent of predicted maximum, respiratory exchange ratios above 1.1, or a plateau in oxygen uptake.

Rather than reducing each person to a single peak number, the researchers dissected the entire response at four physiological checkpoints: rest, the first ventilatory threshold (VT1), the second ventilatory threshold (VT2), and peak effort. VT1 marks the onset of lactate accumulation, when ventilation begins rising to expel the additional carbon dioxide released as bicarbonate buffers the approaching acidosis; VT2 signals the steep second acceleration as the body nears its anaerobic limit. A sports medicine physician identified the thresholds using the modified V-slope method, ventilatory-equivalent curves, the slope of ventilation against carbon dioxide output, and end-tidal carbon dioxide trends, with the respiratory exchange ratio as supporting context. Peak values were extracted from 30-second moving averages of breath-by-breath data and ensemble averages of roughly a dozen heartbeats within the same epochs. Linear mixed-effects models then tested group, stage, and group-by-stage interaction effects for every variable, treating stage as a repeated factor within subjects under an autoregressive covariance structure. The interaction term is the decisive statistic: it asks whether the shape of the response across the whole exercise trajectory differs between groups — a far more demanding test than comparing isolated peaks.

The pulmonary verdict was uniform. Work rate, relative and absolute oxygen uptake, minute ventilation, tidal volume, breathing frequency, the ventilatory equivalents for oxygen and carbon dioxide, carbon dioxide production, end-tidal carbon dioxide pressure, and respiratory exchange ratio all advanced in lockstep across the four stages in both groups, exactly as exercise physiology dictates. Minute ventilation climbed from roughly 12 liters per minute at rest to well over 120 at peak, end-tidal carbon dioxide crested near the first threshold before falling as hyperventilation took over, and the respiratory exchange ratio crossed 1.1 at exhaustion in both groups. Every Stage effect was highly significant; not a single Group effect or Group × Stage interaction reached — or even approached — significance. At maximal effort, the diabetes group sustained an estimated 218 watts against the controls’ 235, ventilated 122 liters of air per minute against 130, and consumed 39.5 milliliters of oxygen per kilogram per minute against 41.3 — margins so narrow the models could not distinguish them from chance.

The cardiovascular data carried the deeper revelation, because stroke volume was tracked directly. Heart rate climbed from the high seventies or low eighties at rest to the mid-180s at peak in both groups, with no detectable difference at any stage. Stroke volume surged from about 83 milliliters at rest to roughly 109–116 by the first ventilatory threshold, then settled into the textbook plateau while heart rate carried the remaining load toward a cardiac output approaching 21 liters per minute. As heart rate accelerated, left ventricular ejection time shortened by roughly 50 milliseconds, compressing the filling window — and the heart compensated, as healthy recreationally active hearts do, by contracting more forcefully to hold stroke volume steady. End-diastolic volume drifted only modestly, ejection fraction rose to the mid-70s before easing slightly, and systemic vascular resistance collapsed from above 2,000 to under 720 indexed units as working muscle demanded perfusion. Oxygen pulse, the gas-exchange surrogate that multiplies stroke volume by the arterial-venous oxygen difference, followed the same path in both groups — textbook physiology, indistinguishable between the arms of the study.

Set against earlier literature, the result marks a clean break. Pooled analyses had reported lower maximal oxygen uptake and reduced oxygen pulse in type 1 diabetes, alongside that suspiciously linear heart-rate curve; a male-only study of uncomplicated patients attributed the pattern to dysregulated cardiac contractility; and a classic 1990s study that did measure stroke volume documented wildly heterogeneous responses in long-term patients, some unable to sustain the early rise at all. The Finnish data sit at the opposite pole: a stroke-volume plateau indistinguishable from controls, heart-rate trajectories without statistical deviation, and a hemodynamic pattern reproducing the classic physiology of the recreationally active heart. The discrepancy, the authors argue, likely reflects cohort composition rather than contradiction — earlier studies folded in people with complications, overweight, recent-onset disease, lower glycemic control, or unreported activity levels. Their result also echoes an earlier Finnish trial in which a year of unsupervised individualized training raised peak oxygen uptake and oxygen pulse equally in patients and controls, despite no improvement in glycemic control — evidence that the capacity to adapt to training is preserved.

The authors are emphatic about what this null result cannot prove. Fifteen participants with diabetes is a small group, and with unequal arms the study had limited power to detect subtle effects; an absence of detectable differences under these conditions is not evidence of physiological equivalence. Blood pressure was not recorded during exercise, recovery metrics fell outside this analysis, and autonomic responses went unmeasured — consequential omissions, because companion studies from the same cohort have already surfaced quiet anomalies: orthostatic testing revealed subtly elevated systolic blood pressure and heightened sympathetic activation in these same volunteers, and a parallel paper found attenuated fast heart-rate recovery after peak effort, hinting at delayed parasympathetic reactivation even in clinically pristine patients. The self-reported activity data carry their own caution: a Mann–Whitney U test flagged a difference in weekly light-intensity activity frequency at p = 0.028, but the Hodges–Lehmann median estimate dissolved once its 95 percent confidence interval embraced zero, and because some participants reported more than one intensity on the same day, the counts were not capped at seven days. Larger studies with prospective power calculations, heart-rate variability measured during exercise, blood pressure monitoring, and continuous glucose metrics such as time in range are the next frontier. Until then, the message for active adults whose diabetes has spared their vessels, nerves, eyes, and kidneys is quietly striking: driven to its physiological ceiling, the uncomplicated diabetic heart performs like anyone else’s.

Tirzepatide is a once-weekly injectable medicine that activates two hormone receptors: those for glucose-dependent insulinotropic polypeptide, or GIP, and glucagon-like peptide-1, or GLP-1. These hormones are released by the intestine after eating and help coordinate the body’s response to food. Tirzepatide enhances insulin secretion when blood glucose is elevated, suppresses the release of glucagon, slows the movement of food through the stomach and reduces appetite. Because its effects on insulin secretion are largely glucose dependent, the drug is designed to lower glucose without continuously forcing the pancreas to release insulin when blood sugar is already low. That mechanism makes it an attractive partner for basal insulin, which provides a steady background supply of insulin but does not always control the sharp glucose rises that follow meals.

Basal insulin is commonly prescribed when tablets and other non-insulin treatments are no longer sufficient. It is usually adjusted upward until fasting glucose reaches a target range, yet this strategy can leave post-meal glucose excursions inadequately controlled. Increasing insulin doses can also cause weight gain and raise the risk of hypoglycemia, particularly when the treatment is intensified without addressing appetite, gastric emptying and glucagon activity. The SURPASS studies tested whether adding tirzepatide could address several of these problems at once. In SURPASS-5, participants receiving basal insulin were assigned to tirzepatide or placebo. In SURPASS-6, tirzepatide was compared with insulin lispro, a rapid-acting mealtime insulin. In both studies, the insulin treatments were titrated toward predefined glucose targets, allowing investigators to compare tirzepatide with active insulin intensification rather than with an unchanged background regimen.

The new report was an exploratory post hoc subgroup analysis, meaning that the researchers examined previously collected trial data after the original studies had been completed. The participants were divided according to baseline HbA1c, age, duration of type 2 diabetes and basal-insulin dosage. HbA1c is a blood marker that reflects average glucose exposure over roughly two to three months; a value above 8.5 percent indicates particularly poor glycemic control in the context of these trials. The analysis therefore asked a practical clinical question: does tirzepatide work mainly in a narrow group of relatively young people with recently diagnosed diabetes, or can it also help patients with long-standing disease, higher starting HbA1c and substantial insulin requirements? The investigators evaluated changes in HbA1c, body weight and hypoglycemia safety across the subgroups.

Across the pooled population, tirzepatide produced marked improvements in glycemic control when added to basal insulin. The reductions in HbA1c were generally consistent across the prespecified categories, including people who began treatment with HbA1c at or below 8.5 percent and those whose levels were higher. This consistency matters because severe insulin resistance and longer disease duration can reduce the body’s remaining capacity to produce insulin. Tirzepatide does not simply provide another source of insulin; it amplifies nutrient-dependent hormone signalling, improves insulin sensitivity indirectly through weight loss and suppresses excessive glucagon. The trial results indicate that these effects remained clinically meaningful even when the underlying diabetes was more advanced or when participants were already receiving higher basal-insulin doses.

The analysis also reinforced the drug’s effects on body weight, an outcome that distinguishes tirzepatide from simply adding more insulin. In the original SURPASS programs, tirzepatide was associated with weight loss, whereas insulin-based intensification generally led to weight gain or little change. That difference is biologically significant. Excess adipose tissue, particularly visceral fat, contributes to insulin resistance through inflammatory signalling and altered lipid metabolism. Weight reduction can lower the amount of insulin required to control glucose and may improve the efficiency of insulin action in muscle and liver. Tirzepatide’s appetite-suppressing and gastric-emptying effects reduce energy intake, while its metabolic actions improve the handling of glucose after meals. The pooled findings suggest that the weight benefit was not restricted to a particular age group or starting insulin dose, although individual responses varied.

Hypoglycemia was another central focus of the study. Any treatment added to insulin must be judged not only by how far it lowers HbA1c but also by whether it drives glucose dangerously low. Tirzepatide’s glucose-dependent mechanism theoretically limits insulin release when glucose concentrations fall, but basal insulin remains capable of causing hypoglycemia. In SURPASS-5 and SURPASS-6, insulin doses were actively adjusted according to protocol, and the pooled analysis assessed episodes of clinically important low blood sugar across the different participant subgroups. The overall safety pattern supported the use of tirzepatide alongside basal insulin without suggesting that the drug created a new, subgroup-specific hypoglycemia signal. Nonetheless, the findings do not eliminate the need for careful insulin titration, glucose monitoring and individualized dose reductions when glucose levels improve rapidly.

The comparison with SURPASS-6 is particularly relevant to everyday treatment decisions. When a patient using basal insulin remains above target, clinicians often add rapid-acting insulin before meals. That approach can be effective, but it requires multiple daily injections, frequent glucose checks and detailed adjustment of meal-time doses. It may also increase the risk of hypoglycemia and promote weight gain. In SURPASS-6, tirzepatide was tested against insulin lispro, providing a direct comparison between a weekly incretin-based therapy and conventional mealtime insulin intensification. The pooled report does not suggest that every patient should abandon prandial insulin, but it adds evidence that a dual GIP/GLP-1 receptor agonist can be a potent alternative for many people who need better control despite basal insulin.

The researchers caution that the analysis has limitations. Because it was conducted after the original trials and was not primarily designed to compare every subgroup, the findings should be interpreted as exploratory rather than as definitive proof that the drug performs identically in all patient categories. The two trials also differed in their comparators, treatment durations and study designs: placebo was used in SURPASS-5, while insulin lispro was used in SURPASS-6. Participants enrolled in randomized clinical trials may also receive more intensive follow-up than people treated in routine practice. In addition, the analysis examined selected baseline characteristics and cannot establish how tirzepatide would perform in every population, including people with severe kidney disease, advanced frailty or type 1 diabetes. Gastrointestinal adverse effects, such as nausea, diarrhea and vomiting, remain relevant considerations for treatment decisions.

Even with those caveats, the findings strengthen the case for tirzepatide as a flexible intensification option for people whose type 2 diabetes remains uncontrolled on basal insulin. The clinical appeal lies in the combination of effects: lower HbA1c, reduced body weight and no obvious loss of hypoglycemia safety across the examined subgroups. The results also highlight a broader shift in diabetes treatment, away from viewing glucose control as a problem solved only by adding more insulin. Modern therapies can target appetite, gut-hormone signalling, glucagon biology and insulin sensitivity simultaneously. For patients and clinicians, the practical question is no longer simply whether insulin should be increased, but which strategy can improve glucose control while minimizing the metabolic costs of treatment. The pooled SURPASS-5 and SURPASS-6 analysis suggests that, for a wide range of adults using basal insulin, tirzepatide may be one of the most powerful answers currently available.

Subject of Research: Tirzepatide as an add-on treatment to basal insulin in adults with type 2 diabetes and inadequate glycemic control.

Article Title: Tirzepatide as an Add-on for Participants with Inadequate Glycemic Control Using Basal Insulin: Pooled Subgroup Analysis of SURPASS-5 and -6

Article References: Bajaj HS, Billings LK, Sharma P, Levine JA, Rodriguez A, Patel H, et al. “Tirzepatide as an Add-on for Participants with Inadequate Glycemic Control Using Basal Insulin: Pooled Subgroup Analysis of SURPASS-5 and -6.” Diabetes Therapy, volume 17, pages 787–797, published 10 April 2026.

Image Credits: AI Generated

Keywords: Tirzepatide, type 2 diabetes, basal insulin, HbA1c, glycemic control, hypoglycemia, weight loss, GIP/GLP-1 receptor agonist, SURPASS-5, SURPASS-6

The research, led by R.G. McCoy, S. Patel, L. Faust and colleagues, addresses a central challenge in modern diabetes care: two people who receive the same diagnosis can face very different futures. One may remain relatively stable with consistent treatment, while another may rapidly develop kidney disease, cardiovascular problems, vision loss or other complications. Conventional risk calculators often provide a single estimate based largely on information collected at one point. Dynamic models, by contrast, can update predictions as new clinical data become available, potentially offering a more responsive picture of individual risk.

That distinction matters because diabetes is not a static illness. Blood glucose measurements fluctuate, treatment regimens are intensified or discontinued, and complications can appear gradually or emerge suddenly. Acute events may include severe metabolic disturbances or urgent cardiovascular episodes, while chronic complications develop over months or years and can damage the kidneys, eyes, nerves, heart and blood vessels. By incorporating information collected during follow-up, a dynamic model can reflect how a patient’s risk changes after diagnosis rather than assuming that the initial clinical profile remains unchanged.

Technically, these models are designed to combine baseline characteristics with time-updated measurements. A patient’s age, diabetes-related laboratory results, blood pressure, kidney function, medication history and coexisting conditions can all contribute to a changing risk estimate. Instead of calculating probability only once, the model repeatedly processes new observations and revises its forecast. This approach is closely related to longitudinal prediction, in which the timing and sequence of clinical events are as important as the values themselves. In practical terms, a model might distinguish between a patient whose risk indicators improve after treatment and one whose measurements deteriorate despite therapy.

The study focuses specifically on adults newly diagnosed with diabetes, a group for whom early decisions may have long-term consequences. At the time of diagnosis, clinicians must determine how intensively to monitor a patient, which therapies to prioritize and whether additional screening is needed. Yet early clinical data can be incomplete or ambiguous. A person may have undetected kidney damage, cardiovascular risk or metabolic instability that becomes visible only after several visits. A prediction system capable of learning from this accumulating information could help clinicians identify which patients need more frequent assessment and which may be safely managed through standard follow-up.

Validation is a critical part of this work. A model can appear impressive when tested on the same kind of data used to develop it, but that performance may not hold when applied to different patients or later clinical records. Development and validation therefore serve separate purposes: the first stage identifies patterns associated with future complications, while the second examines whether those patterns produce reliable predictions beyond the original modeling process. For risk prediction, researchers typically assess discrimination—how well the model separates people at higher and lower risk—and calibration, or how closely predicted probabilities match observed outcomes. Both are essential if a model is to influence real clinical decisions.

The researchers’ emphasis on both acute and chronic complications is especially important because the two categories demand different forms of prevention. Acute complications may require rapid recognition and immediate intervention, meaning that a model must identify short-term changes in risk. Chronic complications, in contrast, are often shaped by cumulative exposure to high glucose, hypertension, inflammation and other biological stresses. Their prevention depends on sustained control and screening over years. A single prediction framework that can address both timescales could offer a more integrated view of diabetes care, linking near-term safety with long-term preservation of organ function.

The potential impact extends beyond individual appointments. Health systems increasingly collect large volumes of electronic health record data, but those data are not automatically transformed into useful clinical guidance. Dynamic prediction models could provide a way to convert repeated laboratory tests, diagnoses and medication changes into structured risk estimates embedded in routine care. Such tools might support reminders for eye or kidney screening, identify people who could benefit from treatment escalation and help allocate specialist resources. They could also support shared decision-making by giving patients a clearer explanation of why monitoring intensity or treatment recommendations change over time.

At the same time, prediction is not the same as certainty, and a risk estimate cannot replace clinical judgment. Models may reflect biases in the populations from which their data were drawn, perform differently across healthcare systems or become less accurate as treatment patterns change. A high predicted risk does not guarantee that a complication will occur, just as a low predicted risk does not eliminate the possibility. Any clinical implementation would require careful evaluation of fairness, transparency, workflow integration and the consequences of false alarms or missed cases. The most useful system would not simply produce a number; it would connect that number to an understandable, evidence-based action.

The broader message from McCoy, Patel, Faust and their colleagues is that diabetes risk assessment may be moving away from the one-time calculator and toward a continuously updated clinical forecast. For adults newly diagnosed with diabetes, this could mean that prevention is guided not only by who they are at diagnosis, but also by how their health changes afterward. If validated across diverse populations and integrated responsibly into care, dynamic prediction could help transform diabetes management from a reactive response to complications into a more anticipatory strategy—one that detects danger earlier, personalizes follow-up and keeps pace with the disease itself.

Subject of Research: Dynamic prediction of acute and chronic diabetes complications among adults newly diagnosed with diabetes

Article Title: Predicting acute and chronic diabetes complications among adults newly diagnosed with diabetes: development and validation of dynamic risk prediction models

Article References: McCoy, R.G., Patel, S., Faust, L. et al. “Predicting acute and chronic diabetes complications among adults newly diagnosed with diabetes: development and validation of dynamic risk prediction models.” Nature Communications (2026). https://doi.org/10.1038/s41467-026-76673-3

Image Credits: AI Generated

DOI: 10.1038/s41467-026-76673-3

Keywords: Diabetes, diabetes complications, risk prediction, dynamic models, acute complications, chronic complications, precision medicine, electronic health records, clinical validation, personalized healthcare