GLP-1's for type one diabetes?

 a recent study found impressive benefit of GLP1 receptor agonists in patients with Type 1 diabetes (T1D), leading to decreases in cardiovascular and kidney outcomes (see dm1 GLP1 dec cardiovasc NatureMed2026 in dropbox, or https://doi.org/10.1038/s41591-026-04274-0)

Details:

-- this study utilized the Optum Labs Data Warehouse, a national dataset including about 300 million patients from >60 health systems

-- the study design was a sequential target trial emulation after propensity matched score weighting (a mouthful, described below in the commentary; this technique is a way to effectively simulate a randomized controlled trial based on existing data from a large database

-- for the study, they found 174,678 individuals with T1D with 6,092,537 person-trials (person-trials includes those patients who were found to start using a GLP-1 at every time period over the course of the study)

-- baseline variables, after weighting of the 2 groups:

    -- mean age 43, 47% female, 66% white/17% Black/9% Hispanic/2% Asian

    -- smoking: never 53%/former 21%/current 21%

    -- obesity: normal wt 10%/overwt 25%/class1 obesity 29%/class2 19%/clases3 17%

    -- eGFR (creatinine-based) <60 in 11%/60-89 in 27%/>90 in 52%; urine albumin/creatinine <30 in 42%/30-300 in 17%/>300 in 6%

    -- Charlson comorbidity index 2.6 (this index is a marker of medical comorbidities where a score of >3 is considered a significant comorbidity burden associated with higher mortality risk)

    -- hypertension 67%, coronary heart disease 16%, stroke 4%, heart failure 7%, dyslipidemia 63%, peripheral arterial dz 2%, GI disease 11%, diabetic retinopathy 7%, cirrhosis 0.5%, chronic liver disease 11%

    -- prior hospitalizations for diabetic ketoacidosis (DKA) 4%, hypoglycemia 2%, acute pancreatitis 2%

    -- A1c 9.0%, duration of DM1 2.9 years

        -- insulin pump use 13%, continuous glucose monitor 26%

    --meds taken:

        -- ACEi 33%/ARB 18%, b-blocker 21%, CCB 14%, diuretics 24%, statins 55%, metformin 52%, SGLT2 17%, sulfonylurea 13%, DPP4 9%, TZD 4%, corticosteroid 12%, antidepressant 23%, non-GLP1 med for obesity 2%

-- primary outcomes: MACE (composite of MI, stroke, or all-cause mortality) and ESRD (end-stage renal disease, encompassing dialysis or kidney transplant)

-- secondary outcomes: heart failure, major adverse liver outcomes (persons with a history of cirrhosis, liver transplantation or liver cancer were excluded, though they included subsequent development of the composite of decompensated cirrhosis, hepatocellular carcinoma or liver transplantation)

-- safety outcomes: hospitalization for DKA, hospitalization for severe hypoglycemia, and composite of biliary disease; pancreatitis, bowel obstruction or gastroparesis

-- of this huge study population, they found 14,488 person-trials of individuals initiating a GLP-1 during the median follow-up of 38 months

-- analysis was through an intention-to-treat assessment of the primary outcomes

Results:

--5-year weighted MACE risk:

    -- initiation of GLP-1: 4.3%

    -- non-initiation of GLP-1: 5.0%

        -- risk difference: −0.7% (−1.2% to −0.2%)

        -- 15% lower risk, with hazards ratio across the study period of 0.85 (0.77–0.95)

-- Risk of ESRD:

    -- initiation of GLP-1: 1.6%

    -- non-initiation of GLP-1: 1.9%

        -- risk difference: −0.3% (−0.6% to 0.0%)

        -- 19% lower risk, HR 0.81 (0.69–0.95)

-- risk of MI:

    -- initiation of GLP-1: 1.9%

    -- non-initiation of GLP-1: 1.8%

        -- risk difference: −0.2% (−0.5% to 0.1%)

        -- 21% lower risk, HR 0.79 (0.66–0.95)

-- risk of all-cause mortality:

    -- initiation of GLP-1: 1.5%

    -- non-initiation of GLP-1: 1.9%

        -- risk difference: −0.4% (−0.7% to 0.0%)

        -- 16% lower risk, HR 0.81 (0.71-1.00)

Figure a above is the cumulative incidence of MACEs; Figure b is cumulative incidence of ESRD

--------------------------------------

secondary outcomes:

-- heart failure:

    -- initiation of GLP-1: 2.4%

    -- non-initiation of GLP-1: 2.7%

        -- risk difference: -0.3% (−0.7% to 0.1%)

        -- 18% lower risk, HR 0.82 (0.71–0.94)

-- major adverse liver events (history of cirrhosis, liver transplantation or liver cancer were excluded, though these adverse liver events below were the subsequent development of the composite of decompensated cirrhosis, hepatocellular carcinoma or liver transplantation)

    -- initiation of GLP-1: 1.4%

    -- non-initiation of GLP-1: 1.8%

        -- risk difference: -0.4% (-0.7% to -0.1%)

        -- 28% lower risk, HR 0.72 (0.60 to 0.85)

-- weight loss of 5%:

    -- initiation of GLP-1: 68.9%

    -- non-initiation of GLP-1: 63.3%

        -- risk difference: 5.6% (​4.6% to 6.7%)

        -- 25% higher likelihood, HR 1.25​ (1.23–1.28)

-- weight loss of 10%

    -- initiation of GLP-1: 45.1%

    -- non-initiation of GLP-1: 40.2%

        -- risk difference: 5.0% (3.7% to 6.0%)

        -- 22% higher likelihood, HR 1.22 (1.19–1.26)

-- weight loss of 15%:

    -- initiation of GLP-1: 27.7%

    -- non-initiation of GLP-1: 24.9%

        -- risk difference: 2.8% (1.7% to 3.8%)

        -- 14% higher likelihood, HR 1.14 (1.09–1.18)

-- no difference in weight loss, or risks of MACE or ESRD if patients were <45yo vs >45yo, or if baseline A1c was <8% vs >8%

-- the researchers also assessed traffic accidents as a negative control in order to assess residual and unmeasured confounding; this analysis revealed no significant difference between GLP-1 initiators or non-initiators in these traffic accidents (the negative control is useful since if there were major comorbidity differences, this might have led to differences in traffic accidents)

safety outcomes:

 -- DKA

    -- initiation of GLP-1: 6.0%

    -- non-initiation of GLP-1: 7.1%

        -- risk difference: -1.1% (−1.6 % to -0.5%)

        -- 17% lower risk, HR 0.83 (0.76–0.90) with GLP-1

-- severe hypoglycemia

    -- initiation of GLP-1: 2.2%

    -- non-initiation of GLP-1: 2.6%

        -- risk difference: -0.4% (−10.7% to -0.0%)

        -- 18% lower risk, HR 0.82 (0.72–0.94) with GLP-1

-- gastrointestinal events (those with history of GI disease were excluded)

    -- initiation of GLP-1: 9.9%

    -- non-initiation of GLP-1: 9.7%

        -- risk difference: 0.2% (−0.6 % to 1.0%)

        -- HR 1.05 (0.99–1.13), a nonsignificant trend to being worse with the GLP-1

Commentary:

-- treatment of type 1 diabetes (T1D) has improved dramatically over the years, in step with improved diabetes technology, and leading to improved A1c values: https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2837376

    -- however only 20% of people in 2020-2021 actually achieved the goal A1c<7%, though this was an increase from about 8% in 2009-2011

 -- 31% of patients with T1D developed major cardiovascular events (MACE), in a Pittsburgh study with 30-year follow-up of 536 patients who were initially free of cardiovascular disease: https://pubmed.ncbi.nlm.nih.gov/34782353/

    -- this study assessed patients with A1c around 10% vs A1c about 8%, finding that those in the higher A1c group had a three-fold increased CVD risk

-- 7% of patients with T1D develop end-stage renal disease (ESRD) with a risk of death 10 times as high as in other patients with T1D, per a Finnish study of 29,906 patients from 1965 to 2011 who were diagnosed with T1D and were <30yo at diagnosis, with median follow-up of 20 years.   https://doi.org/10.2337/dc17-2364

    -- this study found that younger children (those who were diagnosed with T1D less than 5yo), had the least ESRD

    -- however, this study was flawed by the probable ecological fallacy: there was no comment about the achieved A1c in the individuals. did those who actually developed ESRD have very uncontrolled diabetes?? and were those with A1c<7% free from ESRD? we are unable to decipher this with an overall average A1c

-- there were two early very limited, small, short-term studies on GLP-1 use in patients with T1D finding increased risk of symptomatic hypoglycemia and hyperglycemia with ketosis associated with using liraglutide

    -- though a later trial found improved safety with continuous glucose monitoring and a meta-analysis found no increased risk of severe hypoglycemia

-- there are technical issues in developing robust studies documenting a relationship between T1D and subsequent cardiovascular and kidney outcomes:

    -- the T1D population is young, so extended follow-up is needed for this group to detect these long-term outcomes

    -- over a long-term trial, there is a moving target: there are profound differences in lifestyle over time, T1D therapies including many technological and pharmacological advances (eg continuous glucose monitoring, introduction of newer and better meds...), leading to very different achieved A1c levels. so baseline variables are not so useful in determining the much-later endpoints; we need to know about changes over time.

    -- and A1c as a surrogate marker of diabetes control: patients with T1D on insulin often develop insulin resistance, which correlates with insulin dose and is associated with a 17% increased risk of cardiovascular disease over 10 years and a 16% increase in all-cause mortality: https://pubmed.ncbi.nlm.nih.gov/39018337/

-- so, in the above studies, it is certainly plausible that GLP-1 meds might well help in preventing these cardiovascular and renal adverse effects, as found in type 2 diabetes (T2D), by lowering weight, lowering insulin dose required, and overall improving cardiac and renal outcomes perhaps also from decreasing the desire to eat (including eating less not-so-great foods such as ultraprocessed ones, junk foods overall, etc) and decreasing the insulin resistance by lowering the dose of insulin needed.

-- the current trial is a "target trial" emulation, utilizing the huge Optum labs database of 300 million patients. then using propensity score weighting to divide patients into 2 groups with similar baseline characteristics though this weighting, and then assessing subsequent outcomes. this approach thereby emulates a randomized controlled trial finding that "real-world evidence studies can reach similar conclusions as RCTs when design and measurements can be closely emulated":  jamanetwork.com/journals/jama/fullarticle/2804067

-- as with other target emulation studies, the researchers sequentially emulated 135 target trials constructed every consecutive month between January 2011 and March 2024, where individuals initiating GLP-1s within a month of enrollment were classified as initiators vs the noninitiators

    -- both groups could use other noninsulin glucose-lowering agents within a year before the baseline

-- the researchers included secondary outcomes of heart failure and liver disease because there have been associations in the medical literature between GLP1 use and improvements in these outcomes (perhaps in part just by weight loss??)

-- the overall results of this study:

-- significant risk reduction with GLP-1s for MACEs (15%) and ESRD (19%), presumably by lowering the insulin requirement and reducing inflammation, improving insulin sensitivity and metabolic control, enhancing endothelial function and decreasing platelet aggregation

    -- per other trials with type 2 diabetes (T2D), this benefit is likely independent of weight loss (as this current trial had less weight loss than the T2D trials), though the weight loss undoubtedly helps

-- no increased risk of hospitalization for DKA or severe hypoglycemia, consistent with observational studies with semaglutide and tirzepatide

-- no difference in outcomes by age groups or A1c levels at baseline

    -- this study with T1D patients broadens the scope of the prior T2D studies, since there is a much larger percentage of patients with T1D who are younger than in the T2D studies

-- there was also benefit in all of the secondary analyses, including an 18% reduction in hospitalizations for heart failure (which is larger than the 11-13% reduction in those with T2D, perhaps by reducing NAFLD, now referred to as MASLD:  https://pmc.ncbi.nlm.nih.gov/articles/PMC9772979/)

-- there was also a 20% reduction in patients with T1D in adverse severe liver events, also found in those with T2D

Limitations:

-- one major limitation is that we have limited granular information:

    -- no information on which GLP1 was taken and at what dose. the fact that there was no significant increase in GI symptoms suggest a low dose was taken

   -- no information on the amount of time individuals were taking the GLP1s

    -- the information about the comorbidities, and other meds taken were buried in the supplementary material; these data were aggregates, which is also subject to the ecological fallacy: we know that x% of people had hypertension, dyslipidemia, or were on statins, ACEi/ARBs etc etc. But, as a simpler analysis: finding that a town with high consumers of salt had more hypertension prove an association?? what if the few people not eating salt had more hypertension?? the issue is that a community or group average does not necessarily in dictate causality: one needs individual-level data

    -- and there was no information about the doses of the meds (eg insulin, etc), how they changed during the study, the severity of comorbidities or how they changed, the achieved lab values over time (eg, what the individuals' lipid results were over the study), etc                                                           

-- i do think that target trial emulation is a powerful tool when a full-scale RCT is not possible, which is pretty often the case. after all, it is hard to carry out an RCT where one group has a strict diet vs usual diet and assess the outcome 20 years later, or one group had to stop alcohol and the other continued to drink with many-year followup, etc. But it is clear that the target trial emulation quality of the results depends on the extensiveness of the database. if it includes lots of information about medical comorbidites, meds, lab tests, and social issues, and updates these pieces of information periodically during the study, its emulation to an RCT is much more real and interpretable. if not, it raises more of a strong suggestion that the emulation reflects reality

-- by current standards, there was not optimal statin use (only 55%); since end-stage renal disease was an assessed outcome, it would have been more accurate to have cystatin-based eGFR: https://gmodestmedblogs.blogspot.com/2023/12/cystatin-c-better-predictor-of-bad.html

-- as a huge database, there might have been incorrect data inputted. for example, only 25% of the patients were <40yo, and  53% were >50yo. perhaps some of the older people were misclassified as T1D??

-- though there is information about prior hospitalizations for DKA or hypoglycemia, many patients perhaps had less severe DKA or hypoglycemia that may well have been treated outside of the hospital (however this would apply to both groups in the study)

so, i do believe that the current study appropriately suggests a potentially important intervention

-- in my limited experience with GLP-1 receptor agonists in patients with type 1 diabetes, very overweight individuals lose weight, are able to decrease smoking/drinking, require less insulin (presumably from losing weight, but perhaps benefiting from less insulin resistance and thereby likely a lower cardiovascular risk), and have better diabetes control

-- and this target trial emulation technique is really the only way to develop an RCT-type of study dealing with the long-term outcomes from lifestyle targets and informing us on likely clinical outcomes

-- though it is important to have individual-based information to make sure that we are avoiding that problem of ecological fallacy, as per above

geoff

-----------------------------------

If you would like to be on the regular email list for upcoming blogs, please contact me at gmodest@bidmc.harvard.edu

to get access to all of the blogs:  go to http://gmodestmedblogs.blogspot.com/ to see the blogs in reverse chronological order

to get access to the dropbox: https://www.dropbox.com/scl/fo/pcf4vymtpo77f49wd2ljr/ANyJqB40CuBd-DMEr67XDoI?rlkey=5dsalmuhf8ekd0svsgvn93xm1&dl=0

or you can just click on the magnifying glass on top right, then type in a name in the search box and get all the blogs with that name in them