Introduction

Diabetes mellitus is a prevalent comorbidity in patients with community-acquired pneumonia (CAP). It is associated with adverse outcomes of CAP, such as increased mortality [1–5], prolonged length of hospital stay [2, 6], more severe illness [3, 6, 7] and an increased rate of complications [3, 6]. Similar to a general CAP population, higher CAP severity scores (e.g. PSI [pneumonia severity index], CURB65 [confusion, blood urea nitrogen, respiratory rate, low blood pressure, age ≥ 65 years] and A-DROP [age, dehydration, respiratory failure, orientation disturbance, low blood pressure] [1–3, 8–10]), time to antibiotics, early complications, advanced age, multilobar infiltrates and comorbidities predict death in a diabetic cohort with CAP [5, 11]. Diabetes-specific risk factors predicting outcome include persistent hyperglycaemia during hospitalisation and presence of long-term diabetes complications like retinopathy and nephropathy [3, 10, 12]. In diabetic patients with any infectious disease, HbA1c level and diabetes duration have been shown to be additional diabetes-specific predictors for infection-related death [5, 13]. Data for these risk factors in patients with CAP in particular are missing.

Interestingly, the predictive value of initial serum glucose for outcome of CAP in diabetic patients is debated [1, 5, 12, 14]. Nonetheless, in non-diabetic individuals with CAP, hyperglycaemia upon admittance is an unambiguous, stress-related predictor for adverse outcome [1, 12, 15].

We have recently shown that adjunct corticosteroids as compared with placebo in a general CAP population shortens the time to reach clinical stability (time to clinical stability, TTCS) and duration of hospital stay [16, 17]. Herein, our aim was to evaluate whether pre-existing diabetes and/or hyperglycaemia had an influence on the effect of corticosteroids on TTCS and other endpoints in this well-defined cohort of patients with CAP from a previous multicentre study.

Methods

Study design and participants

This is a preplanned subanalysis of a prospective randomised, double-blind, placebo-controlled multicentre trial, in which the additive effect of prednisone in CAP was investigated. The main study demonstrated the benefit of corticosteroids in CAP, with a reduction in TTCS and duration of hospital stay [17].

Study details have been published [18]. Briefly, upon admission to hospital, consecutive patients with CAP were randomised to receive 50 mg of prednisone for 7 days or to receive placebo. Inclusion criteria were age 18 years or older and hospital admission with CAP [19]. Exclusion criteria were permanent inability for informed consent, active intravenous drug use, acute burn injury, gastrointestinal bleeding within the past 3 months, known adrenal insufficiency, a condition requiring more than 0.5 mg/kg per day prednisone equivalent, pregnancy, breastfeeding and severe immunosuppression [17].

The primary objective of this analysis was to investigate whether the effect of corticosteroids on TTCS is modified by the presence of diabetes mellitus. Diabetes mellitus was defined as having pre-existing diabetes, ascertained by medical records.

Secondary objectives were as follows: (1) effect modification by the presence of initial hyperglycaemia, defined as initial (first) blood glucose measurement > 11.0 mmol/l on day 1 (i.e. before start of study medication), irrespective of diabetes diagnosis or administration of glucose-lowering agents and (2) the effect of corticosteroids on blood glucose values during the time of study and their influence on outcome. The primary endpoint for primary and secondary objectives was TTCS, defined as time to stabilisation of vital signs at two consecutive measurements ≥ 12 h apart. Secondary endpoints included time to effective hospital discharge, all-cause mortality, total and intravenous duration of antibiotic treatment, CAP complications (including recurrence, acute respiratory distress syndrome, empyema, nosocomial infections until day 30, severe adverse events possibly related to CAP, intensive care unit [ICU] stay, re-admission to hospital) and new insulin requirement at day 30.

The conduct of the trial adhered to the declaration of Helsinki and Good Clinical Practice Guidelines, and ethical committees of all participating hospitals approved the study before patient recruitment. The trial was registered with ClinicalTrials.gov (registration no. NCT00973154).

Procedures

Informed consent was obtained within 24 h of admission to hospital; thereafter, patients started receiving study medication. Study nurses assessed patients for clinical stability (defined as normalisation of body temperature, oxygen saturation, blood pressure and heart rate) [17] every 12 h during hospital stay. All patients were treated according to published CAP guidelines [20]. Baseline data included medical history, relevant comorbidities, clinical variables relating to pneumonia and all variables required for the calculation of the PSI [21]. In diabetic individuals, the type of diabetes, type of on-going treatment (oral, insulin, diet only) and HbA1c levels (in 60% of diabetic patients) were assessed on admission. In all patients, glucose measurements were performed at 08:00, 12:00, 18:00 and 20:00 hours on days 1, 3, 5 and 7 to screen for corticosteroid-induced hyperglycaemia. All blood glucose measurements were based on fingerstick tests and explicitly included at least one glucose measurement scheduled 2 h postprandially. The doctors and nurses administering treatments were instructed to perform daily glucose measurements if the screening glucose levels showed relevant hyperglycaemia and to consider starting or increasing anti-hyperglycaemic treatment if indicated. Structured follow-up telephone interviews for secondary outcomes after discharge were done on day 30.

Statistical analysis

To analyse the effect of corticosteroids in diabetic patients and patients with hyperglycaemia, the per-protocol population was subdivided as follows: (1) diabetic patients and non-diabetic patients and (2) patients with and without initial hyperglycaemia, irrespective of diabetes diagnosis. For the primary endpoint, we performed Cox regression models for TTCS and included interaction terms. Significant results in interaction analysis provide evidence for effect modification. Therefore, interaction p values ≥ 0.05 indicate no significant difference regarding the effect of prednisone in the respective subgroup compared with the entire study population, while p < 0.05 suggests that the subgroup differs from the entire study population regarding the effect of prednisone. Secondary endpoints were compared between the two arms by the Mann–Whitney test for continuous data or linear regression, χ2 tests or Fisher’s exact test, respectively, or binomial/Poisson regression for binary/count data, and the logrank test or Cox regression for time-to-event data. For all endpoints, estimates of the effect size and corresponding 95% CIs were provided.

The effect of corticosteroids on glucose levels was assessed by comparing the following variables between the prednisone and placebo group: (1) mean glucose value during study time; (2) SD of glucose values as a surrogate marker for glycaemic variability during study time and (3) difference in outcome according to glucose values. The mean glucose value during the study period was calculated for every patient using the mean of all available blood glucose values measured during the study time. For every patient, we calculated the SD of the mean glucose value during the study time as a surrogate marker for glucose variability [22, 23].

Further, to investigate the influence of glycaemia on outcome, we analysed the influence of higher mean glucose value and higher glycaemic variability on outcomes in different subgroups (e.g. patients receiving prednisone vs placebo; diabetic vs non-diabetic patients). First, we adjusted these data for PSI. Second, we adjusted for PSI and additional insulin requirement.

Results

Baseline characteristics

In this study, 802 eligible patients were randomly assigned to receive either prednisone or placebo (Fig. 1). After blinded post-randomisation exclusion of protocol violators and patients retrospectively not meeting eligibility criteria, the per-protocol population consisted of 362 treated with prednisone and 365 patients receiving placebo. These groups were further subdivided into patients with and without pre-existing diabetes mellitus. In the non-diabetic placebo group, one patient had to be excluded from the analysis due to lack of blood glucose measurements.

Fig. 1
figure 1

Patient flow by treatment group

Full size image

Table 1 shows the baseline characteristics of diabetic and non-diabetic patients in the treatment subgroups. The mean age of diabetic patients was 75 years, and 73% were men. Sixty-seven per cent of diabetic patients were in high-risk PSI classes IV and V, as compared with 45% of non-diabetic patients, resulting in a higher PSI score in diabetic patients. A greater proportion of diabetic patients had comorbidities compared with non-diabetic patients, with a significantly higher rate of heart failure and renal insufficiency. Within the diabetic group, 32% of the patients in the prednisone and 33% in the placebo group had insulin pre-treatment. Mean initial glucose was 8.5 mmol/l in diabetic patients and 6.8 mmol/l in non-diabetic patients. None of the patients treated with prednisone experienced diabetic ketoacidosis.

Table 1 Baseline characteristics and clinical variables of participants (per-protocol population)
Full size table

Effect of corticosteroids on outcome in diabetic vs non-diabetic patients

First, we investigated TTCS in both groups, with HR > 1.0 corresponding to faster TTCS. Adjunct prednisone shortened TTCS in both diabetic and non-diabetic individuals, as shown in Table 2 (HR [95% CI] 1.65 [1.16, 2.35] vs 1.30 [1.10, 1.53]), with no evidence for effect modification in interaction analysis (p = 0.44). With regards to all-cause mortality at day 30, there was no significant difference with adjunct prednisone between diabetic and non-diabetic patients (p for interaction = 0.065; OR for diabetic patients 0.34 [95% CI 0.07, 1.77], p = 0.18; OR for non-diabetic patients 2.02 [95% CI 0.80, 5.08], p = 0.14). We also performed a severity-adjusted analysis (adjusted for PSI), which showed similar effects for mortality (diabetic patients, OR 0.18 [95% CI 0.03, 1.27], p = 0.09); non-diabetic patients, OR 1.84 [95% CI 0.69, 4.88], p = 0.22; p for interaction = 0.052).

Table 2 Overview of primary and secondary endpoints: diabetic vs non-diabetic patients
Full size table

Time to effective hospital discharge, total duration of antibiotic treatment and the incidence of CAP complications did not differ between the treatment groups, either in diabetic or in non-diabetic patients. Also, duration of intravenous antibiotic treatment was reduced in non-diabetic patients, with no evidence for interaction. In diabetic patients treated with prednisone, four patients had new insulin requirement at day 30. They were all pre-treated with oral glucose-lowering agents; their respective HbA1c values at study inclusion ranged from 7.0% to 11.4% in the two different subgroups. No patients in the non-diabetic cohort treated with prednisone, but one previously non-diabetic patient in the placebo group, had new insulin requirement at day 30.

Effect of corticosteroids on outcome in patients with vs without initial hyperglycaemia

In both patients with and without hyperglycaemia on admission, TTCS was shorter in the prednisone group than in the placebo group without evidence for effect modification (ESM Table 1, p for interaction = 0.14). There were no significant differences in other secondary endpoints, although differences in mortality could not be compared due to the small number of patients in the subgroups.

Effect of corticosteroids on glucose levels

Patients in the prednisone group had higher mean glucose levels and higher glycaemic variability during the study time as compared with patients in the placebo group (Table 3). This was found in both diabetic and non-diabetic patients, although diabetic patients had overall higher mean glucose levels. Furthermore, in the prednisone subgroup, more patients experienced hyperglycaemia during the study time (88% of diabetic and 52% of non-diabetic patients). Only 2.9% of the diabetic patients remained normoglycaemic during the study time with no significant difference between the prednisone and placebo group (p = 0.35). In non-diabetic patients, the proportion of normoglycaemic patients was significantly lower in the prednisone group (7.3%) than in the placebo group (29.7%, p < 0.001).

Table 3 Effect of corticosteroids on glucose levels
Full size table

Influence of corticosteroids on in-hospital insulin use

In the group with diabetes, not previously treated with insulin, 76% of patients treated with prednisone required new insulin treatment during study time as compared with 67% of diabetic patients treated with placebo (p = 0.16) (Table 2). Within the non-diabetic subgroup, 20% of patients treated with prednisone and 14% of patients treated with placebo required new insulin therapy (p = 0.001). The interaction analysis showed no statistical significance (p = 0.16). Diabetic patients treated with prednisone did not require more additional insulin than diabetic patients treated with placebo, contrary to the findings in non-diabetic patients, in whom treatment with prednisone led to more additional insulin treatment (see ESM Table 2 for details).

Effect of glucose levels and glucose variability on outcomes

We also investigated the association of overall mean glucose values and overall glucose variability with outcomes stratified by prednisone treatment. In unadjusted analysis, higher mean blood glucose was associated with longer TTCS only in the placebo group (OR 0.91 [95% CI 0.85, 0.98], p = 0.012), but not in the prednisone group (OR 0.98 [95% CI 0.92, 1.03], p = 0.38). Similarly, in patients with higher mean glucose levels, those treated with placebo had a longer hospital stay than those treated with prednisone (coefficient for placebo 0.63 [95% CI 0.24, 1.02], p = 0.002 vs, coefficient for prednisone 0.29 [95% CI −0.003, 0.58], p = 0.053). However, after adjusting for PSI and for supplemental insulin, higher overall glucose levels did not influence the primary and secondary endpoints (Table 4 and ESM Table 3).

Table 4 Influence of higher overall glucose level on outcome
Full size table

Higher glycaemic variability was associated with lower mortality in the prednisone group (PSI-adjusted OR 0.44 [95% CI 0.22, 0.88], p = 0.02), Table 5). The effect on mortality remained significant after additional adjustment for insulin (OR adjusted for PSI and supplemental insulin 0.44 [95% CI 0.22, 0.90], p = 0.03). We repeated these analyses for the diabetic and the non-diabetic subgroups (see ESM Table 3 for higher mean glucose levels during study time and ESM Table 4 for higher glycaemic variability). Higher glycaemic variability in diabetic patients correlated with longer time to effective hospital discharge. This effect did not persist after adjusting for insulin treatment. Additionally, we found that in PSI-adjusted analysis, higher glycaemic variability was associated with reduced TTCS in non-diabetic patients and with lower mortality in diabetic patients, including both treatment groups. Both effects remained significant after additional adjustment for insulin.

Table 5 Influence of higher glycaemic variability on outcomea
Full size table

Discussion

Our study has two key findings. First, the benefit of adjunct prednisone treatment in CAP patients with diabetes was similar across the overall population. This was valid for the primary endpoint TTCS, as well as for the secondary endpoints, without an increase in complications. Even though 76% of diabetic patients in the prednisone group required new insulin treatment during study time, as compared with 67% in the placebo group, only four diabetic patients treated with prednisone (6%) required new insulin treatment at day 30. In at least three of these patients, this was not related to the corticosteroid medication itself but rather to the unsatisfactory metabolic status already existing before and at admission. We therefore consider the risk of diabetes progression due to a short-term corticosteroid intervention in the diabetic population to be negligible. This is in agreement with the few publications available on this topic, where complications of corticosteroid treatment are mainly seen during long-term treatment rather than in a short-term protocol of 7 days [24, 25]. Regarding mortality in the primary analysis, diabetic patients derived a greater benefit from prednisone treatment than non-diabetic patients, although this finding was not significant. Diabetic patients had a higher CAP severity. It is known that patients with a high CAP severity treated with corticosteroids experience less treatment failure [26] and show a better mortality reduction than patients with mild disease [27, 28]. However, our finding cannot be related to CAP severity only, since results remained similar after adjustment for PSI.

The second main finding of our study was that overall mean glucose levels during hospitalisation and glucose variability were increased by prednisone treatment. Due to the hyperglycaemic effects of prednisone, there is concern about the impact of corticosteroid-induced hyperglycaemia on outcome. Accordingly, patients in the placebo group with a higher mean glucose and/or variability had a longer TTCS and prolonged time to effective hospital discharge. However, this effect did not persist after adjustment for disease severity and for supplemental insulin. Thus, hyperglycaemia in the overall population is likely a surrogate marker for disease severity and, hence, for outcome.

Previous studies investigating outcome prediction of admission and overall glucose levels in diabetic patients have shown similar results. Two studies showed that serum glucose levels on admission had no influence on outcome in diabetic CAP patients [1, 12] but that after adjusting for disease severity, persistent hyperglycaemia was an independent factor for worse outcome of CAP [12].

Contrarily, hyperglycaemia in patients treated with prednisone in our study (irrespective of diabetes status) did not have a negative effect on either primary or secondary endpoints.

Importantly, in diabetic patients, higher blood glucose values—reflecting a side-effect of prednisone—were not associated with a negative outcome. This finding persisted after adjustment for supplemental insulin. Possibly, diabetic patients are less sensitive to acute hyperglycaemic states and are hence protected from the toxic effects of high blood glucose levels [12]. Therefore, diabetic patients with CAP are an appropriate target group for adjunct prednisone treatment. Prednisone has a strong anti-inflammatory effect, modulating the immune system on a broad base towards restitution. Not only do corticosteroids play a crucial role in resolution of inflammation, they also enable an effective response to bacterial infections and tissue damage [29]. This is especially important in severe courses of CAP [30], which occur more often in diabetic patients.

The following limitations of the study have to be mentioned. First, as a secondary subgroup analysis, it may be underpowered for some comparisons due to smaller patient numbers in subgroups. Second, diagnosis of diabetes was based on medical records. There might have been undiagnosed diabetic patients potentially confounding the results. However, when we analysed subgroups of initial hyperglycaemia and normoglycaemia, we found the same results as in the subgroups of diabetic and non-diabetic patients. Third, we included patients hospitalised with CAP of any severity and patients with severe CAP were under-represented. Fourth, according to in-hospital guidelines, all patients with hyperglycaemia were treated. Based on our data, it is therefore not possible to conclude whether corticosteroid-induced hyperglycaemia should be treated or not.

In summary, our results show that the beneficial effect of adjunct prednisone in CAP on outcome was also valid for individuals with diabetes or hyperglycaemia upon hospital admission. Hyperglycaemia in diabetic patients or due to adjunct prednisone did not have a negative effect on outcome. For this reason, we advocate the use of corticosteroids as an appropriate additive treatment for CAP in patients with diabetes or hyperglycaemia on admission.