Benefits and Harms of Sodium-Glucose Co-Transporter 2 Inhibitors in Patients with Type 2 Diabetes: A Systematic Review and Meta-Analysis

1Patients with type 2 diabetes are characterized by hyperglycaemia with elevated levels of glycated haemoglobin A1c (HbA1c) which may lead to microvascular and macrovascular disease. Between 2012 and 2014, three sodium-glucose co-transporter 2 inhibitors (SGLT2-i), canagliflozin, dapagliflozin and empagliflozin, were approved by the US Food and Drug Administration (FDA) and the European Medicines Agency (EMA) for the treatment of patients with type 2 diabetes. SGLT2-i inhibit glucose reabsorption in the proximal tubules of the kidneys, increasing urinary glucose excretion and reducing the amount of circulating glucose. SGLT2-i have been assessed as monotherapy or combined with other antidiabetic agents including metformin, sulphonylureas (SU), dipeptidyl peptidase 4 inhibitors (DPP-4-i), thiazolidinediones (pioglitazone) or insulin.

2In 2015 the American Diabetes Association (ADA) and the European Association for the Study of Diabetes (EASD) recommend SGLT2-i as second-line agents in the management of type 2 diabetes. A recent randomised controlled trial (RCT) evaluated the effect of empagliflozin on cardiovascular disease (CVD)-associated events in 7,020 patients with type 2 diabetes and a high risk of CVD events. The study found that empagliflozin reduced the relative risk of the CVD events including death from cardiovascular causes, non-fatal myocardial infarction and non-fatal stroke by 14% (absolute risk reduction of 1.6%) compared to placebo. Whether the effect is specific for empagliflozin or represents a class effect for SGLT2-i will be assessed in on-going RCTs assessing the effect of canagliflozin. and dapagliflozin on CVD in patients with type 2 diabetes. However, the efficacy and safety of SGLT2-i in patients with a low to moderate cardiovascular risk or in a real world setting, where patients often have multiple co-morbidities and are treated with several drugs, have not been established.

3In contrast to previous meta-analyses evaluating the effects of SGLT2-i in type 2 diabetes, we only included trials, which used the recommended maximum daily doses of the SGLT2-i as we expect these dosages to be the most widely used in the clinic. Lower or higher doses of SGLT2-i might overestimate or underestimate the efficacy or the risk of adverse events. The present approach provides the evidenced-based clinician with a clear and balanced summary of the existing evidence.

4In addition, three studies found that intensive glucose lowering treatments may harm some patients and recently, the safety of SGLT2i was put into question by the regulatory agencies.

5We conducted the present systematic review with meta-analyses of RCTs evaluating the safety and efficacy of the SGLT2-i canagliflozin, dapagliflozin, and empagliflozin administered in highest clinically relevant doses for at least 12 weeks compared to placebo or OAD.

6We conducted our review based on a published protocol (PROSPERO CRD42014008960; S2 File) and adhered to the PRISMA standards for the conduct and reporting of this systematic review and meta-analysis (PRISMA checklist; S3 File).

7Electronic searches were performed in the Cochrane Library, MEDLINE, EMBASE, the Science Citation Index and the WHO Trial Search Database, using the following search string: “((Sodium glucose (All Fields) AND co-transporter (All Fields)) OR (2-(3-(4-ethoxybenzyl)-4-chlorophenyl)-6-hydroxymethyltetrahydro-2H-pyran-3,4,5-triol (Supplementary Concept) OR 2-(3-(4-ethoxybenzyl)-4-chlorophenyl)-6-hydroxymethyltetrahydro-2H-pyran-3,4,5-triol (All Fields) OR dapagliflozin (All Fields)) OR (canagliflozin (Supplementary Concept) OR canagliflozin (All Fields)) OR (empagliflozin (Supplementary Concept) OR empagliflozin (All Fields))”. Additional manual searches were performed in reference lists of relevant papers. We obtained additional data on e.g. heart rate, ALT and lipids from the study investigators, the manufacturers and the YODA-project (details listed in S1 File). The last search update was October 2015.

8We included English-language, full paper, double-blind RCTs conducted in adult patients (at least 18 years of age) with type 2 diabetes. The interventions assessed were the recommended daily target doses of the SGLT2-i canagliflozin 300 mg; dapagliflozin 10 mg; empagliflozin 25 mg. Controls could receive placebo or OAD including metformin, SU or DPP-4-i. We only included RCTs with a treatment duration of at least 12 weeks. Co-interventions (‘add-on’ therapies) with other antidiabetic agents were allowed if administered to both the intervention and control groups. We excluded studies, which involved participants with impaired kidney function and SGLT-2i only approved in Japan (ipragliglozin, luseogliflozin, tofogliflozin) or in clinical development (ertugliflozin, remogliflozin, sotagliflozin).

9Trial selection was carried out by two review authors (HS and CB) who independently reviewed the search results and selected trials for inclusion, with involvement of a third review author (CB or TV) if necessary to resolve disagreements. Multiple publications, which reported results from the same RCT, were grouped into ‘studies’ (S1 File).

10Our primary outcomes were HbA1c (change from baseline) and serious adverse events defined as the number of participants experiencing cancer (all cancers, bladder cancer, breast cancer), death, severe hypoglycaemia, ketoacidosis and CVD. The secondary outcomes were fasting plasma glucose (FPG) (mmol/L), change in body weight (kg), systolic and diastolic blood pressure (SBP and DBP (mmHg)), heart rate (beats per minute (bpm)), plasma lipid profile (low-density lipoprotein (LDL) cholesterol (mmol/L) (which is known to increase the risk of CVD), high-density lipoprotein (HDL) cholesterol (mmol/L) and triglyceride (mmol/L)), alanine amino transferase (U/L), adverse events leading to discontinuation and drug-related adverse events. We also evaluated non-serious adverse events defined as the number of participants experiencing urinary tract infections (UTI), genital tract infections (GTI); 'non-severe' hypoglycaemia, and serum creatinine.

11Trial characteristics (methods, participants, interventions, study outcomes, potential risks of bias, and funding source) were recorded. Three authors (HS, MFG and MBC) independently identified outcomes from each included study and extracted outcome data into extraction forms (Excel spreadsheets). Consensus was reached through discussion. For trials presenting data from more than one treatment period (e.g. 26 and 52 weeks), data from the longest treatment period were used. For studies with multiple treatment arms for example SGLT2-i, other OAD and placebo. We conducted separate evaluations and analyses of a) SGLT2-i versus placebo and b) SGLT2-i versus other OAD.

12The bias risk assessment followed the Cochrane Collaboration’s risk of bias assessment tool. In each domain, studies were given a rating of low, unclear or high risk. We used the Grades of Recommendation, Assessment, Development and Evaluation (GRADE) system to describe the quality of the evidence and the strength of recommendation, 'high' to 'very low’.

13We undertook meta-analyses in RevMan using random-effects models, unless stated otherwise. We chose the random-effects model due to an expected heterogeneity. We conducted the analyses with the assumption that if the estimates were similar, then any small-study effects had little effect on the intervention effect estimate. If the random-effects estimates were more beneficial, we planned to re-evaluate whether it was reasonable to conclude that the intervention was more effective in the smaller studies. However, in all of our analyses, the conclusions of the fixed-effect and random-effects meta-analyses were consistent. Based on the expected clinical heterogeneity, we expected that our analyses would display statistical between-trial heterogeneity (I2 > 0%). For random-effects models, precision will decrease with increasing heterogeneity and confidence intervals will widen correspondingly. We therefore (a priori) planned to report the random-effects model under the assumption that they would provide the most conservative (and a more correct) estimate of the intervention effect. We present results as mean differences (MD) or relative risks (RR) with 95% confidence intervals (CI). For effect sizes of MD, values greater than 0.70 were treated as large; values between 0.40 and 0.70 as moderate; and values less than 0.40 but greater than 0.10 as small. We conducted subgroup analyses on the basis of SGLT2-i type (canagliflozin, dapagliflozin, empagliflozin), and on the basis of the type of OAD (metformin, SU, or DPP-4-i). Differences between subgroups were reported using tests for subgroup differences expressed as P values. I2 values were used as a measure of heterogeneity and are reported if they exceeded 30%. For meta-analyses with at least 10 RCTs, publication bias and other small study effects were assessed in regression analyses and funnel plots. For continuous variables, linear regression of the intervention effect estimates on their standard errors, weighting by 1/(variance of the intervention effect estimate), was used (Egger test). For dichotomous outcomes Z/sqrt(V) was regressed against sqrt(V) (Harbord test), where Z is the efficient score and V is Fisher's information (the variance of Z under the null hypothesis).

14We identified 1,087 potentially eligible records through our searches and included 42 RCTs described in 59 published reports (Fig 1). The total number of participants was 24,500 (S1 File). Thirty-four RCTs compared SGLT2-i versus placebo and 12 compared SGLT2-i versus OAD. Four RCTs were multi-arm, comparing SGLT2-i versus placebo and AD.

15Thirty-four RCTs compared SGLT2-i versus placebo. Seven RCTs evaluated canagliflozin 300 mg, 17 evaluated dapagliflozin 10 mg, and 10 evaluated empagliflozin 25 mg (Table 1). Twelve RCTs compared SGLT2-i versus OAD (Table 1). Of these 12 trials, four compared canagliflozin versus glimepiride or sitagliptin and four compared dapagliflozin versus metformin, glipizide or saxagliptin. The remaining four studies compared empagliflozin versus linagliptin, glimepiride or sitagliptin. The maximum doses of metformin were 2000 mg or 1500 mg. The doses of the other OADs was 1 to 8 mg for glimepiride, 20 mg for glipizide, 100 mg for sitagliptin, 5 mg for saxagliptin and 5 mg for linagliptin.

16Thirty-one RCTs were multicentre and multinational carried out in USA, Europe and Asia and three RCTs were conducted Japan. The duration of the RCTs ranged from 12 weeks to 102, or 104 weeks, with the longest duration being 208 weeks.

17We excluded 17 RCTs (S1 File) for the following reasons: the dose used in the RCTs did not meet our criteria, open label extension with optional cross-over of placebo, included patients with kidney disease, was not double blind or assessed the combination of SGLT2-i and OAD or insulin. We did not include any abstracts or RCTs published in other languages than English.

18All RCTs had a low risk of bias in the assessment of randomisation (allocation sequence generation and concealment) and were double blind. One RCT was classified as unclear risk of attrition bias. The published trial report stated that “Approximately 93% of the patients in each treatment arm completed the 24-week double-blind treatment period”. The description of the statistical analyses explained that patients were excluded from the analyses if they did not receive the intervention or did not have follow up assessments. We classified three RCTs as unclear or high risk of reporting bias. One RCT did not provide a clear description of secondary/exploratory outcome measures. The second RCT listed the glomerular filtration rate as the only primary outcome in the registered trial protocol, but in the trial publication, primary outcomes included renal function, blood pressure, and circulating plasma volume. The third RCT did not provide information about adverse events. All RCTs were industry-funded and were classified as unclear risk of bias in the domain ‘other biases’. Accordingly, none of the trials had a low risk of bias in all domains.

19Random-effects meta-analysis of 34 RCTs with 9,154 patients showed that SGLT2-i were associated with a beneficial effect on HbA1c compared with placebo (MD -0.69%, CI -0.75 to -0.62%, Fig 2). Between study heterogeneity was detected (I2 = 75%) and we found evidence of small study effects in regression analysis (P = 0.015) and visual inspection of a funnel plot. In addition, subgroup analysis showed a clear difference between subgroups (test for subgroup differences P = 0.008). The largest effect size was seen for canagliflozin (-0.85%, -0.99 to -0.71%; Fig 2).

20Analyses of 12 RCTs showed that SGLT2-i were associated with a larger reduction in HbA1c than OAD (-0.20%, -0.28–0.13%; Fig 3). There was between study heterogeneity, evidence of small study effects (P = 0.0385), and no difference between subgroups of trials stratified by the OAD (P = 0.11). We found no difference in HbA1c-reduction between SGLT2-i and metformin (-0.05%, 0.21 to 0.12%, Fig 3), but a larger HbA1c reducing effect of SGLT2-i compared with SU (-0.15%, -0.21 to -0.08%) and DPP-4-i (-0.25%, -0.36 to -0.14%).

21Only a few serious adverse events were recorded and no differences were seen between SGLT2-i versus placebo (RR 0.99, CI 0.87 to 1.12, 34 RCTs, 10,703 patients) or OAD (1.02, 0.78 to 1.34, 12 RCTs, 6,759 patients). Five patients randomized to SGLT2-i and six patients randomized to placebo reported severe hypoglycaemia (0.75, 0.23 to 2.43, n = 5,077 patients). In trials comparing SGLT2i versus SU, no patients versus three patients experienced a severe hypoglycaemic event (0.13, 0.02 to 0.73, n = 814). No cases of ketoacidosis were reported. In total, 32 of 3,201 patients allocated to SGLT2-i and 29 of 3,223 allocated to placebo developed cancers (1.04, 0.6 to 1.83; 19 RCTs). Only one case of bladder cancer was reported, in the placebo arm of a dapagliflozin study. Six of 2,767 patients were diagnosed with breast cancer in the SGLT2-i arms compared with two of 2,789 patients in the placebo arms (1.73, 0.56 to 5.36; 18 RCTs). When analysing RCTs comparing SGLT2-i with other OAD, seven patients allocated to canagliflozin and three allocated to sitagliptin were diagnosed with other types of cancer than bladder or breast cancer (2.41, 0.69 to 8.37; 2 RCTs). One patient allocated to canagliflozin developed breast cancer and none developed bladder cancer.

22CVD events were recorded in 56 of 5,438 patients randomized to SGLT2-i versus 45 of 5,263 randomized to placebo (1.24, 0.86 to 1.81) or OAD (0.78, 0.27 to 2.32).

23As shown in Table 2, analysis of 33 RCTs with 8,914 patients found that FPG levels were 0.9 mmol/L lower in the SGLT2-i arm compared with the placebo arm (-1.0 to -0.8 mmol/L). There was no small study effect (P = 0.122) and a difference between subgroups (P = 0.04). The largest effect size was seen for canagliflozin (Table 2).

24We found no difference between SGLT2-i and metformin or SU but a beneficial effect compared with DPP-4-i (-1.0, 1.3 to 0.7 mmol/L, Table 3). The between trial heterogeneity was moderate to high in all analyses.

25SGLT2-i were associated with a loss of body weight compared with placebo (-2.1 kg, -2.3 to -2.0 kg). The effect was different in subgroups stratified by the type of SGLT2-i (P < 0.01) with the largest weight reduction associated with canagliflozin (Table 2). SGLT2-i also reduced the body weight compared to OAD (Table 3).

26SGLT2-i reduced the systolic blood pressure compared with placebo (-3.9 mmHg, -4.6 to -3.3 mmHg), there were subgroup differences (P = 0.03), with the largest effect seen for canagliflozin (Table 2). SGLT2-i also reduced the systolic blood pressure compared with OAD (Table 3). A similar effect was seen in analyses of the diastolic blood pressure (Tables 2 and 3). The heart rate did not differ between patients allocated to SGLT2-i versus placebo (-0.6 bpm, -1.3 to 0.0 bpm) (Table 3). However, there was a difference between subgroups when compared with placebo (P = 0.04) and empagliflozin induced a modest increase in heart rate (Table 2). The heart rate in the SGLT2-i group was lower than in the DPP-4-i group (-1.50 bpm, 2.7 to 0.4 bpm).

27SGLT2-i was associated with increased HDL cholesterol compared with placebo (0.05 mmol/L, 0.04 to 0.07 mmol/L). A similar result was achieved for LDL cholesterol (0.09 mmol/L, 0.04 to 0.14 mmol/L), whereas triglyceride decreased (-0.09 mmol/L, -0.16 to -0.02 mmol/L). Subgroup analysis showed a difference between subgroups, with the largest effects seen for canagliflozin on HDL cholesterol, LDL cholesterol and triglycerides (Table 2). SGLT2-i increased HDL and LDL cholesterol, but did not reduce triglycerides compared to OAD (SU and DPP-4-i) (Table 3).

28Analyses of 18 RCTs with 3,719 patients found evidence that SGLT2-i reduced alanine aminotransferase levels compared with placebo (-2.8 U/L, CI -4.0 to -1.7 U/L) or OAD (Table 3).

29STLG2-i were associated with a 0.60 μmol/L increase in creatinine compared with placebo (0.1 to 1.1 μmol/L) (Table 2). The largest increase was seen for canagliflozin. Analysis of SGLT2-i versus other OAD showed no difference between SGLT2-i and metformin or DPP-4-i (Table 3).

30Compared with placebo, SGLT2-i were associated with an increased risk of UTI (1.14, 1.0 to 1.3) and GTI (4.34, 3.35 to 5.63). SGLT2-i were also associated with an increased risk of UTI compared with metformin (2.01, 1.01, 3.98), but not SU (1.05, 0.84 to 1.31) or DPP-4-i (0.89, 0.67 to 1.19). SGLT2-i were associated with an increased risk of GTI compared with metformin (4.48, 1.76 to 11.42), SU (5.41, 3.64 to 8.03) and DPP-4-i (3.69, 2.42 to 5.63; P < 0.00001).

31An analysis of 33 RCTs with 10,440 patients found fewer episodes of non-severe hypoglycaemia in the placebo group compared to the SGLT2-i group (1.11, 1.03 to 1.2). Subgroup analysis showed a difference between subgroups (P = 0.04). The largest risk of hypoglycaemia was seen for canagliflozin (1.53, 1.15 to 2.03). Dapagliflozin (1.07, 0.95 to 1.19) and empagliflozin (1.03, 0.9 to 1.19) did not increase the risk of non-severe hypoglycaemia. SGLT2-i were associated with a decreased risk of non-severe hypoglycaemia compared with SU (0.16, 0.11, 0.22), but not compared with metformin (0.5, 0.18 to 1.43) or DPP-4-i (1.00, 0.49 to 2.02). In the SGLT2-i group, more participants experienced drug-related adverse effects (1.45, 1.27 to 1.66) and discontinued treatment (1.28, 1.08 to 1.51) compared with placebo.

32We gave evidence from RCT data a high quality rating, but downgraded it if there was unexplained clinically important heterogeneity, the study methodology had a risk of bias, the evidence was indirect, there was important uncertainty around the estimate of effect, or there was evidence for reporting bias. Therefore, it was possible for RCT data to have a very low quality of evidence if several of these concerns were present. Where we downgraded the evidence, it was mainly because there was risk of bias, small study effects, or considerable heterogeneity. Some outcomes had relatively few events (e.g. mortality) and wide CIs (imprecision). The results of many meta-analyses had moderate to high levels of statistical heterogeneity (inconsistency). The heterogeneity between the trials resulted from differences between the three SGLT2-i and in the outcome measures reported, the duration of follow up and the trials inclusion criteria. In the assessment of the primary outcomes, we downgraded the quality of the evidence for glycated haemoglobin in the analyses comparing SGLT2-I by two levels to low quality, due to heterogeneity and evidence of publication bias or other small study effects. We also downgraded the outcome serious adverse events and analyses comparing SGLT2-i versus OAD to moderate quality evidence due to uncertainty (wide confidence intervals) and heterogeneity, respectively.

33The highest approved doses of canagliflozin, dapagliflozin and empagliflozin compared with placebo, were effective in reducing HbA1c in patients with type 2 diabetes. In spite of the large number of RCTs with a low risk of bias in several domains, we downgraded the evidence to low quality. Based on our assessment of publication bias and other smalls study effects, we found evidence of bias and therefore a risk that the analyses overestimate the intervention benefit. In the included RCTs, SGLT2-i had no discernible beneficial or harmful effects on serious adverse events including mortality, cancer, ketoacidosis, severe hypoglycaemia, bladder cancer, breast cancer or other cancer types. SGLT2-i also had no effect on CVD events, but SGLT2-i were associated a beneficial effect on CVD-associated risk factors including body weight, blood pressure and lipids (although elevations in LDL lipids may be a concern). As expected, SGLT2-i increased the risk of non-serious adverse events, including serum creatinine levels, UTI and GTI. Additional meta-analyses showed similar effects, when comparing SGLT2-i versus other OAD, but the analyses with active comparators included a smaller number of trials and patients. We also identified important potential limitations, which mainly included a high degree of inconsistency. The inconsistency is likely to reflect clinical heterogeneity in terms of the interventions, populations and follow-up times. Furthermore, selective reporting of outcomes (e.g. CVD, cancer etc.) may also bias the estimates. Therefore, it is possible that the true effect differs somewhat from the estimated effects.

34We found statistically clear differences between SGLT2-i in subgroup analyses. The largest effect was seen for canagliflozin in the analyses of HbA1c and CVD-related risk factors. However, none of the trials compared the individual SGLT2-is and the results, therefore, remain exploratory. Thus, the lack of head-to-head comparisons between the SGLT2-i means that we cannot exclude the possibility that the difference between SGLT2-i reflect patient inclusion criteria rather than a true difference between intervention effects.

35Patients with type 2 diabetes have a high risk of adverse CVD outcomes. The effects of SGLT2-i on cardiovascular mortality and morbidity in patients with type 2 diabetes are unknown. In one study, empagliflozin was associated with a lower rate of cardiovascular events compared with placebo. Despite a sample size of more than 24,500 patients in this review, few RCTs reported CVD as an outcome. In our analyses of CVD events, we found no differences between SGLT2-i and placebo or OAD. We only found beneficial effect on outcomes that may be associated with a lower risk of CVD.

36We found a beneficial effect of SGLT2-i on alanine aminotransferase, which is associated with non-alcoholic fatty liver disease in the early phase. Increasing evidence suggests that non-alcoholic fatty liver disease may increase the risk of CVD. SGLT2-i decreased alanine aminotransferase both in comparison to placebo and OAD. While such improvements may be attributed solely to weight loss, rather than drug-specific effects additional evidence is needed to determine the potential clinical implications of the findings.

37We included creatinine, which may reflect dehydration due to the glycosuria. On SGLT2-i, approximately 500 ml of water after treatment is initiated. The loss generally decreases during long term treatment. Increased serum creatinine may although reflect a worsening of kidney function which is predictive of CVD. The largest increase in creatinine levels was found in RCTs evaluating canagliflozin. Whether this translates to an increased risk of CVD events in patients taking SGLT2-i over the long-term is unclear.

38Recently, ketoacidosis has been reported as an adverse effect of SGLT2-i. The RCTs in this review did not routinely report ketoacidosis as an outcome. Theoretically, there is a potential for developing ketoacidosis as a result of the insulin-independent glucose excretion combined with increased glucagon levels. However, a recent large RCT has found a low incidence of ketoacidosis (≤ 0.1%) and that the risk was similar in patients treated with empagliflozin and placebo.

39SGLT2-i are widely studied and several reviews and meta-analyses have recently been published. Compared to these studies our systematic review with meta-analysis has distinct differences in the dosages and outcomes that we address. Zaccardi et al. performed a network meta-analysis that focused on efficacy and safety of SGLT2-i. In contrast to our meta-analysis, they included trials with several different doses of canagliflozin, dapagliflozin and empagliflozin and they reported fewer secondary outcomes than us (we also include e.g. ALT, Creatinine and heart rate). In another meta-analysis, Wu et al. examined the effects of SGLT2-i on cardiovascular events, death and major safety outcomes in adults with type 2 diabetes. No beneficial effects of SGLT2-i were reported. We analysed both efficacy and safety data. In the network meta-analysis by Shyangdan et al., the primary aim was to compare the efficacy of SGLT2-i. The investigators only included trials on SGLT2-i in monotherapy or as add on to metformin in patients with type 2 diabetes. Only a total of 10 trials were included and no data on adverse events were provided.

40Future RCTs would ideally be long-lasting and large-scale comparing SGLT2-i with placebo or existing therapies. Such RCTs should additionally include reporting of serious adverse events such as CVD risk, ketoacidosis and severe hypoglycaemia, and monitoring of renal safety, with adequate follow-up (over one year), to establish the long-term consequences of SGLT2-i therapy.

41Based on our review we found evidence that clinically relevant doses i.e. the recommended daily target doses of SGLT2-i that are included in this review, during more than 12 weeks reduce HbA1c levels in patients with type 2 diabetes compared with placebo and other existing oral therapies. We planned to include high-quality RCTs with clinically relevant doses and sufficient follow up to generate an estimate based on the best available evidence. However, our analyses showed evidence of bias and heterogeneity. Likewise, the incidence of serious adverse events including mortality, CVD and cancer was not increased as a result of SGLT2-i, but reporting was inconsistent. Several CVD risk factors such as obesity, blood pressure and HDL cholesterol may be improved by SGLT2-i therapy, whereas the incidences of UTI and GTI are increased in the SGLT2-i groups. Additional evidence may therefore be needed to determine the benefit and safely of SGLT2-i. The RCTs included in our review were largely carried out in research hospital settings. Given the high prevalence of type 2 diabetes in the general population, RCTs conducted outside the hospital settings seem warranted.