Introduction

Despite advances in the efficacy of, and access to, antibiotic therapies, pneumonia remains the most common global cause of infection-related mortality.1 Incidence of, and mortality from, pneumonia is higher in patients with type 2 diabetes (T2D)2; with the risk further amplified in patients with poor glycaemic control.3 Paradoxically, several glucose-lowering therapies, including sulphonylureas and dipeptidyl peptidase-4 inhibitors (DPP-4i), have not demonstrated benefit in reducing incidence of, or mortality from, pneumonia,4 while long-term thiazolidinedione treatment (eg, pioglitazone) was associated with a modestly increased risk of pneumonia in a meta-analysis of randomised controlled trials (RCTs).5 Therefore, improvement in glycaemic control in isolation may be inadequate in reducing pneumonia risk in patients with T2D, and more specific drug-class specific mechanisms maybe implicated.

Current pharmacological treatment of T2D has been revolutionised with the introduction of two relatively new drug classes: sodium-glucose cotransporter-2 inhibitors (SGLT2i) and glucagon-like peptide-1 receptor agonists (GLP-1 RAs). Both drug classes exert clinical effects far beyond glucose-lowering, including weight loss and protection against cardiorenal complications.6 7 These pleiotropic action of SGLT2i and GLP-1 RA may also contribute to reduced pneumonia risk, in addition to weight-loss induced anti-inflammatory properties.8 9 Meta-analysis of 26 SGLT2i RCTs demonstrated a 13% reduced risk of pneumonia, and a 35% reduced risk of septic shock,10 with a meta-analysis of cohort studies demonstrating a 38% risk reduction.11 Evidence regarding incident pneumonia in patients prescribed GLP-1 RA is less clear. The findings of one study suggests that delayed gastric emptying following GLP-1 RA may increase risk of aspiration pneumonia,12 despite head-to-head comparison of SGLT2i and GLP-1 RA demonstrating no significant difference in hospitalisation secondary to pneumonia,13 and meta-analysis of GLP-1 RA cardiovascular outcome trials suggesting a similar risk of respiratory tract infections.14

The reassuring RCT evidence lacks generalisability to real-world populations, detailing drug efficacy instead of effectiveness. Evidence generated from large cohort studies is considered complementary to RCTs, and given the contradictory evidence from population-based studies to date, and the scarcity of real-world studies assessing impact of GLP-1 RAs on incident pneumonia, further research is needed in this area. In view of the above, the aims of our study were to assess the impact of SGLT2i, and GLP-1 RA, on incident pneumonia risk, and separately to assess their impact on incident severe sepsis risk.

MethodsStudy design

We conducted a cohort study with anonymised data from TriNetX (TriNetX, Cambridge, MA, USA), a global federated health research network that has access to both inpatient and outpatient electronic medical records from healthcare organisations internationally; primarily secondary, and tertiary care providers in North America and Western Europe. This analysis was conducted on the Global Collaborative Network, which contains data from over 135 million patients with access to diagnoses, procedures, medications, laboratory values and genomic information worldwide. The data used in this study was collected on 18 April 2024. Further details have been described elsewhere.15

Study population 1: SGLT2i

We identified all adult patients (>18 years) in TriNetX with a history of T2D, excluding other causes of diabetes (identified through one of: (a) International Classification of Disease (ICD)-10 codes on their electronic medical records; (b) biochemistry (HbA1c>6.4%); or (c) other glucose-lowering therapy (aside from GLP-1 RAs and SGLT2i, which are used for other indications, such as treatment of obesity and heart failure, respectively)), who were prescribed an SGLT2i (treatment), or DPP-4i (reference), at baseline (figure 1).

Figure 1

Flow diagram of SGLT2i versus DPP-4i, and GLP-1 RA versus DPP-4i, cohort selection. DPP-4i, dipeptidyl peptidase-4 inhibitors; GLP-1 RAs, glucagon-like peptide-1 receptor agonists; SGLT2i, sodium-glucose cotransporter-2 inhibitors; T1D, type 1 diabetes; T2D, type 2 diabetes.

Study population 2: GLP-1 RA

We identified all adult patients (>18 years) in TriNetX with a history of T2D, excluding other causes of diabetes, who were prescribed a GLP-1 RA (treatment), or DPP4i (reference), at baseline (figure 1).

For both cohorts, the treatment arm excluded any patient treated with DPP-4i (reference arm) at any point, and visa-versa. DPP-4i were chosen as the reference arm as GLP-1 RAs, SGLT2i and DPP-4i are typically used as second, or third, line glucose-lowering therapy in patients with T2D in the UK, and globally, thus reducing confounding by indication. Moreover, DPP4i are consistently used as reference glucose-lowering therapy in existing literature owing to their lack of pleiotropic action, beyond glycaemic control.11

Index event

The index event followed an active comparator, new user design, where analysis was of new starters of each drug, 2 years after initiation. This lag was introduced to allow for an adequate time for each drug to achieve meaningful efficacy. Patients were followed up for 12 months in an intention-to-treat analysis. The index event must have been after the first diagnosis of T2D to avoid immortal-time bias, and there was a look-back period set to ‘anytime’ within the TriNetX database to ensure detection of all coding of the study drugs, and ascertainment of covariates. Covariate information at baseline was obtained prior to the index event, rather than diagnosis of T2D.

Propensity score matching

Cohorts were propensity score matched (PSM), in a 1:1 ratio, for age, sex, ethnicity, body mass index (BMI), HbA1c, glomerular filtration rate (GFR), presence of hypertension, dyslipidaemia, diseases of the respiratory system, including specifically pneumonia, ischaemic heart disease (as a proxy of smoking) and cancer, as well as prescription of other glucose-lowering therapy, antibiotics, immunosuppressive agents (including biological agents, exogenous adrenal steroids and chemotherapy) and respiratory tract medicines. Definitions are presented in online supplemental table 1.

Outcomes

The primary outcome was time-to-incident pneumonia (ICD-10 J13-J18). The secondary outcome was time-to-incident severe sepsis (ICD10 R65). Use of the term ‘severe sepsis’ reflects ICD-10 coding used for the outcome in this study, accepting that it is not a medical term with clinical utility or relevance, with the simple term ‘sepsis’ more widely adopted. Treatment and reference arms were followed up until the first coding of pneumonia, and separately severe sepsis, on their electronic medical records, within 12 months. Patients who did not develop incident pneumonia were censored from the: (1) end of the time window for analysis or (2) patient’s last known fact date.

Secondary analysis

We assessed whether the impact of SGLT2i, and GLP-1 RA, treatment on incident pneumonia risk differed depending on which glucose-lowering therapy was used as the reference. We additionally identified all adult patients in TriNetX with a history of T2D, excluding other causes of diabetes, who were prescribed one of (1) metformin, (2) sulfonylureas, (3) thiazolidinediones, (4) SGLT2i/GLP-1 RAs (head-to-head analysis). Prescription of the reference arm was excluded from the treatment arm and visa-versa.

Statistical analysis

Statistical analysis for cohort data was performed in situ within the TriNetX platform. Normally distributed baseline characteristics are presented as mean and SD. PSM was performed using logistic regression. TriNetX uses ‘greedy nearest-neighbour matching’ with a calliper of 0.1 pooled SD and difference between propensity scores <0.1. We assessed covariate balance between groups using the standardised mean difference (SMD). SMD<0.1 was considered well matched. The control arm of alternative oral glucose-lowering therapy was considered the reference cohort (HR=1.0) when compared against the treatment arm of patients prescribed an SGLT2i or GLP-1 RA. Survival analysis was performed to estimate the probability of an outcome, at daily time intervals, over 12 months from the index event. An HR and log-rank test were generated. TriNetX uses the R Survival package V.3.2-3. Additionally, for sensitivity analysis, we (1) repeated all analyses with longitudinal repeat coding for each glucose-lowering therapy to ensure drug adherence, (2) repeated analysis when the diagnosis of T2D was within 1 year of the index event to account for diabetes duration, (3) repeated analysis with a lag of 1 year, (4) performed falsification analysis to assess whether there was a difference in incidence of a theoretically unrelated outcome (delusional disorder (ICD-10 F22)) in patients treated with SGLT2i versus GLP-1 RA and (5) calculated E values, representing the minimum strength of association on the HR scale that an unmeasured confounder would need to have with both the exposure (SGLT2i and GLP-1 RA) and the outcome, conditional on the measured confounders, to explain away the observed association; HR+√[HR×(HR−1)].16 The Strengthening the Reporting of Observational Studies in Epidemiology guidelines were followed in the reporting of this prospective cohort study.17

ResultsMain cohort 1: SGLT2i

Our study included 407 792 new patients prescribed SGLT2i, and 573 865 new patients prescribed DPP-4i, prior to propensity score matching (figure 1). Before matching, compared with new users of DPP-4i, new users of SGLT2i were younger, more likely to be white, male, living with obesity and have prevalent hypertension, dyslipidaemia, IHD, diseases of the respiratory system and neoplasms, as well as being coprescribed more glucose-lowering therapy, adrenal corticosteroids and respiratory tract medicine. However, they did have better renal function, demonstrated by their estimated GFR (table 1). After PSM, each cohort was reduced in size and deemed well matched, and included 352 687 patients in both the SGLT2i and DPP-4i arms.

Table 1

Baseline characteristics of participants, from SGLT2i and reference arms, before and after propensity score matching

Survival analysis

Treatment with SGLT2i was associated with a significantly reduced risk of incident pneumonia (HR 0.75 (95% CI 0.73, 0.78)), and severe sepsis (0.75 (0.73, 0.77)), compared with treatment with DPP-4i (table 2, online supplemental table 2 and figure 2). The incidence rate for pneumonia per 1000 person-years in the SGLT2i arm was 25.6 (vs 34.6 in the DPP-4i arm), and for severe sepsis was 20.6 (vs 28.0).

Figure 2

Forest plots demonstrating the HRs for primary outcomes in main and secondary analyses for (a) pneumonia, and (b) severe sepsis. DPP-4i, dipeptidyl peptidase-4 inhibitors; GLP-1 RAs, glucagon-like peptide-1 receptor agonists; SGLT2i, sodium-glucose cotransporter-2 inhibitors.

Table 2

HRs demonstrating incident pneumonia risk in the various main and subgroup analyses

Secondary analysis

Following survival analysis, SGLT2i treatment was associated with a reduced risk of incident pneumonia compared with all other glucose-lowering therapies (1) metformin (0.89 (0.86, 0.92)), (2) sulfonylureas (0.75 (0.73, 0.77)), (3) thiazolidinediones (0.82 (0.79, 0.86)), except for GLP-1 RAs (1.27 (1.23, 1.32)) (table 2). Additionally, SGLT2i treatment was associated with a reduced risk of incident severe sepsis compared with (1) metformin (0.89 (0.86, 0.93)) and (2) sulfonylureas (0.76 (0.74, 0.78)), but SGLT2i may be less effective than thiazolidinediones (1.06 (1.01, 1.10)), and GLP-1 RAs (1.27 (1.22, 1.31)) (online supplemental table 2).

Sensitivity analysis

SGLT2i treatment remained associated with risk reduction against pneumonia, and severe sepsis, compared with DPP-4i, when: (1) repeating coding was used (pneumonia (0.70 (0.66, 0.73)), and severe sepsis (0.75 (0.70, 0.80))); (2) treatment initiation was within 1 year of the diagnosis of T2D (pneumonia (0.75 (0.72, 0.78)) and severe sepsis (0.82 (0.79, 0.86))); and (3) the index event was 1 year post initiation of treatment or reference drug (pneumonia (0.84 (0.81, 0.87)) and severe sepsis (0.84 (0.81, 0.87))) (online supplemental table 3). During falsification analysis, there was no significant difference in the associated incidence of delusional disorder between patients treated with SGLT2i and GLP-1 RA (0.87 (0.71, 1.07)).

Main cohort 2: GLP-1 RAs

Our study included 516 849 new patients prescribed GLP-1 RAs, and 637 820 new patients prescribed DPP-4i, prior to propensity score matching (figure 1). Before matching, compared with new users of DPP-4 inhibitors, new users of GLP-1 RAs were younger, more likely to be white, female, living with obesity and have prevalent hypertension, dyslipidaemia, diseases of the respiratory system and neoplasms, as well as being co-prescribed more glucose-lowering therapy, adrenal corticosteroids, respiratory tract medicine, immunosuppressants and antibiotics. However, their renal function was better (table 3). After PSM, each cohort was reduced and deemed well-matched and included 331 863 patients in both the GLP-1 RA and DPP-4i arms.

Table 3

Baseline characteristics of participants, from GLP-1 RA and reference arms, before and after propensity score matching

Survival analysis

Treatment with GLP-1 RAs was associated with a significantly reduced risk of incident pneumonia (0.60 (0.58, 0.62)), and severe sepsis (0.61 (0.59, 0.63)), compared with treatment with DPP-4i (table 2, (online supplemental table 2 and figure 2). The incidence rate for pneumonia per 1000 person-years in the GLP-1 RA arm was 17.8 (vs 29.3 in the DPP-4i arm), and for severe sepsis was 15.1 (vs 24.6).

Secondary analysis

Following survival analysis, GLP-1 RA treatment was associated with reduced risk of incident pneumonia compared with all other glucose-lowering therapies (1) metformin (0.69 (0.67, 0.72)), (2) sulphonylureas (0.56 (0.54, 0.58)), (3) thiazolidinediones (0.67 (0.64, 0.70) and (4) SGLT2i (0.78 (0.75, 0.80)) (table 2). Additionally, GLP-1 RA treatment was associated with reduced risk of incident severe sepsis compared with (1) metformin (0.71 (0.68, 0.74)), (2) sulphonylureas (0.59 (0.57, 0.60)), (3) thiazolidinediones (0.89 (0.85, 0.93)) and (4) SGLT2i (0.78 (0.75, 0.81)) (online supplemental table 3).

Sensitivity analysis

GLP-1 RA treatment remained associated with risk reduction against pneumonia, and severe sepsis, compared with DPP-4i, when: (1) repeating coding was used (pneumonia (0.59 (0.56, 0.63)), and severe sepsis (0.70 (0.65, 0.75))); (2) treatment initiation was within 1 year of the diagnosis of T2D (pneumonia (0.60 (0.57, 0.63)) and severe sepsis (0.60 (0.57, 0.63))); and (3) the index event was 1 year post initiation of treatment or reference drug (pneumonia (0.79 (0.76, 0.82)) and severe sepsis (0.80 (0.77, 0.84))) (online supplemental table 3).

Discussion

We have performed the largest cohort study in patients with T2D, that assesses the impact of either SGLT2i and/or GLP-1 RA, compared against all other glucose-lowering therapy, on incident pneumonia and severe sepsis risk, to date. We demonstrate that GLP-1 RAs, and to a lesser extent SGLT2i, are associated with a reduced incidence of pneumonia and severe sepsis compared with all other glucose-lowering therapies.

Comparison against existing SGLT2i studies

Despite early clear evidence of the cardiorenal benefits of SGLT2i treatment in patients with T2D,6 previous meta-analysis has demonstrated an increased risk of genitourinary infections, likely related to increased bacterial proliferation resulting from glycosuria.18 Given that SGLT2i have affinity for SGLT1 receptors found in the lungs,19 early hypotheses raised concerns regarding an analogous mechanism in the respiratory tract; with increased glucose levels in respiratory tract mucosa potentially resulting in bacterial proliferation and pneumonia. However, pooled results of randomised, placebo-controlled trials of SGLT2i, that included patients with T2D, and reported incident pneumonia, suggested that SGLT2i reduced the risk of incident pneumonia by 13%.10 When interpreting these results, it is important to consider that the RCTs included were not adequately powered to detect all infection-related outcomes, including only selected patients that may limit generalisability to the general population; for example, these populations were predominantly Caucasian males. The underpowering is highlighted considering this effect size compared with that seen in a meta-analysis of four cohort studies, suggesting a much greater benefit on incident pneumonia risk with SGLT2i (vs DPP-4i) being reduced by 38%.11 However, even using population-based data, underpowering of studies may impact results, as 75% of included studies analysed data from less than 10 000 patients prescribed SGLT2i. The only study to include more than 10 000 patients prescribed SGLT2i was performed using data from South Korea, and hence may not be representative of outcomes in Western populations.20 Reassuringly, the effect size of our current study in patients prescribed SGLT2i vs DPP-4i was 25%, performed in the largest cohort analysed to date.

Comparison of GLP-1 RA studies

The association between GLP-1 RA treatment and incident pneumonia is less clear. To the best of our knowledge, this is the first real-world study to assess the impact of GLP-1 RA against all other oral glucose-lowering therapies on this outcome. One previous meta-analysis of 28 RCTs, involving 77 485 participants, demonstrated a 14% lower risk of all respiratory disease, but no significant difference in respect to pneumonia.21 Similarly, meta-analysis of six RCTs found no significant difference in upper or lower respiratory tract infections in patients treated with GLP-1 RA compared with placebo.14 Both meta-analyses may have been limited by underpowering of pneumonia as an incident outcome due to small sample sizes of the included RCTs. Additionally, we have identified one small retrospective analysis that assessed the impact of GLP-1 RA (vs SGLT2i) on incident pneumonia, which demonstrated no significant difference in pneumonia risk in patients with T2D, however, they did not compare incident pneumonia risk in patients prescribed GLP-1 RA against any other oral glucose-lowering therapy, and given their small sample size, results again may be affected by underpowering.13 Interestingly, a recent study, also using data from TriNetX, did demonstrate that in patients with GLP-1-treated T2D who had undergone endoscopy, their incidence of subsequent aspiration pneumonia was 33% higher.12 However, the risk was only present who had undergone propofol-assisted endoscopy, and is therefore more likely related to impaired airway protection reflexes.22

Biological plausibility

Although we did not, and cannot, examine the mechanistic explanations in our study, the exact mechanistic basis through which the reduced risk is derived remains unclear. The pleiotropic properties of both SGLT2i and GLP-1 RA could, in part, explain the lower incident pneumonia risk. This may be mediated through the associated weight loss following treatment with both agents23–25: mean weight loss of 2 kg with SGLT2i and 10% with the GLP-1 RA, semaglutide, with the main effect being on fat mass.26

The reduced adiposity would have a profound attenuation of chronic, low-grade inflammation.27 Indeed, several pro-inflammatory cytokines are reduced following SGLT2i,28 and GLP-1 RA,29 treatment. More broadly, incident pneumonia and severe sepsis are greatly impacted by comorbid disease, specifically cardiovascular disease,30 and nephropathy,31 and both SGLT2i and GLP-1 RA offer significant cardiorenal protection.6 7 Other specific pathways involved in SGLT2i-induced amelioration of incident pneumonia may include optimisation of haematocrit and haemoglobin, which can improve oxygen delivery to the lung parenchyma.32 In respect to GLP-1 RA specifically, the expression of GLP-1 receptors appears higher in the lung compared with other extra-pancreatic organs, and this may contribute to reduced airway inflammation in patients with chronic obstructive airway disease who are more susceptible to pneumonia.33 Moreover, in vivo models have demonstrated that metformin reduces glucose flux into the lung by reducing transepithelial permeability, independently of its known blood glucose-lowering effects. It is possible that, given the incretin effects of GLP-1 RAs, there may exist beneficial effects of insulin in the lung epithelial glucose permeability.34 Despite these suggestions, further mechanistic evaluation is warranted in this area.

Clinical implications

While patients with T2D are at increased risk of incident pneumonia,2 coupled with increased risk of genitourinary tract infections in patients treated with SGLT2i,18 the results of our study are reassuring to clinicians and patients alike in respect to pneumonia risk following treatment of T2D with SGLT-2i, suggesting there is no analogous mechanism occurring in the respiratory tract as is seen in the genitourinary system. When using DPP-4i as the comparator glucose-lowering therapy, the results of our study may suggest that the number needed to prevent one incident pneumonia in patients with T2D prescribed SGLT2i is 111, and for GLP-1 RA is 91. To put this into starker context, in older adults, the number needed to prevent one incident pneumonia, regardless of a diagnosis of T2D, when prescribed a 13-valent pneumococcal conjugate vaccine (vs not being vaccinated) is 213.35

Limitations

We must acknowledge the limitations to our work. First, these are real-world data, and comparisons are not randomised or controlled. Second, resulting from data being extracted from electronic health records of an administrative database, there is potential for a lack of data completeness. For example, data may not be recorded by the HCO, such as the dose, or duration, of treatment, or recorded in free text that we are unable to extrapolate. We attempted to mitigate against challenges faced in determining treatment duration by ensuring that repeat coding for all glucose-lowering therapies assessed was present and separated by at least 1 year in a sensitivity analysis. Moreover, residual bias confounding is possible, as with any large database study. For example, confounding by indication. In both the USA, and Europe, SGLT2i are recommended specifically for patients with high cardiovascular risk, or with established cardiorenal complications (such as coronary artery disease, congestive heart failure and chronic kidney disease), while GLP-1 RAs are used in patients with obesity and elevated atherosclerotic cardiovascular disease risk. These clinical comorbidities/risk factors that may influence a clinician’s decision to select prescribe these classes of agents over other glucose-lowering therapies (including DPP4 inhibitors) and may mean the use is often indicated for patients with more severe disease/diabetes-related complications. Specifically, in our study, in several of the GLP-1 RA analyses performed, even after thorough adjustment for covariates, the GLP-1 RA arm had minor residual imbalances in respect to BMI (as DPP4i are not used for treatment of obesity). Given the significantly reduced risk of incident pneumonia in all the GLP-1 RA analyses regardless of the higher baseline BMI, we suggest the actual effect size in truly BMI-matched individuals may be greater. Any other unmeasured confounding because of treatment indication for SGLT2i/GLP-1 RA (vs DPP4i) would also likely result in an under-estimate of the true risk reduction of pneumonia in these patients, thereby strengthening our findings. Further, it was not possible to match for healthcare utilisation, or account for drug switching during follow-up. ICD-10 codes, used for identification of outcome disease in the TriNetX platform, do not allow stratification of sepsis by (organ-system) source of infection, and therefore we could not exclude urosepsis in the overall outcome. However, the treatment arm excluded the reference arm drug, at any time, and visa-versa, so patients cannot switch between the specific drugs of interest. There is no clear evidence from meta-analysis of clinical trials or observational studies that SGLT2i increase the risk of urinary tract infections or urosepsis.36 Moreover, the risk of severe sepsis was, in fact, lower in patients prescribed SGLT2i, compared with DPP-4i, and therefore, our results may be a conservative protective benefit. Additionally, we attempted to reduce any risk of unidentified residual confounding through calculation of E values as a quantitative bias analysis to assist readers in the interpretation of the strength of our results.16 Finally, although ICD-10 revision coding is a validated method for identifying pneumonia, variability in diagnostic and coding practices might influence its accuracy.