Sodium glucose cotransporter 2 inhibitors and risk of serious adverse events: nationwide register based cohort studyBMJ 2018; 363 doi: https://doi.org/10.1136/bmj.k4365 (Published 14 November 2018) Cite this as: BMJ 2018;363:k4365
- Peter Ueda, postdoctoral researcher1,
- Henrik Svanström, senior researcher1 2,
- Mads Melbye, professor2 3 4,
- Björn Eliasson, professor5,
- Ann-Marie Svensson, research advisor6,
- Stefan Franzén, senior statistician6,
- Soffia Gudbjörnsdottir, professor5 6,
- Kristian Hveem, professor7 8,
- Christian Jonasson, senior researcher7 8,
- Björn Pasternak, senior researcher1 2
- 1Clinical Epidemiology Division, Department of Medicine, Solna, Eugeniahemmet, T2, Karolinska University Hospital, 17176 Stockholm, Sweden
- 2Department of Epidemiology Research, Statens Serum Institut, Copenhagen, Denmark
- 3Department of Clinical Medicine, University of Copenhagen, Copenhagen, Denmark
- 4Department of Medicine, Stanford University School of Medicine, Stanford, CA, USA
- 5Department of Molecular and Clinical Medicine, Institute of Medicine, University of Gothenburg, Gothenburg, Sweden
- 6The Swedish National Diabetes Register, Västra Götalandsregionen, Gothenburg, Sweden
- 7KG Jebsen Center for Genetic Epidemiology, Department of Public Health and Nursing, Faculty of Medicine and Health Science, Norwegian University of Science and Technology, Trondheim, Norway
- 8HUNT Research Center, Faculty of Medicine, Norwegian University of Science and Technology, Levanger, Norway
- Correspondence to: P Ueda
- Accepted 9 October 2018
Objective To assess the association between the use of sodium glucose cotransporter 2 (SGLT2) inhibitors and seven serious adverse events of current concern.
Design Register based cohort study.
Setting Sweden and Denmark from July 2013 to December 2016.
Participants A propensity score matched cohort of 17 213 new users of SGLT2 inhibitors (dapagliflozin, 61%; empagliflozin, 38%; canagliflozin, 1%) and 17 213 new users of the active comparator, glucagon-like peptide 1 (GLP1) receptor agonists.
Main outcome measures The primary outcomes were lower limb amputation, bone fracture, diabetic ketoacidosis, acute kidney injury, serious urinary tract infection, venous thromboembolism, and acute pancreatitis, as identified from hospital records. Hazard ratios and 95% confidence intervals were estimated by using Cox proportional hazards models.
Results Use of SGLT2 inhibitors, as compared with GLP1 receptor agonists, was associated with an increased risk of lower limb amputation (incidence rate 2.7 v 1.1 events per 1000 person years, hazard ratio 2.32, 95% confidence interval 1.37 to 3.91) and diabetic ketoacidosis (1.3 v 0.6, 2.14, 1.01 to 4.52) but not with bone fracture (15.4 v 13.9, 1.11, 0.93 to 1.33), acute kidney injury (2.3 v 3.2, 0.69, 0.45 to 1.05), serious urinary tract infection (5.4 v 6.0, 0.89, 0.67 to 1.19), venous thromboembolism (4.2 v 4.1, 0.99, 0.71 to 1.38) or acute pancreatitis (1.3 v 1.2, 1.16, 0.64 to 2.12).
Conclusions In this analysis of nationwide registers from two countries, use of SGLT2 inhibitors, as compared with GLP1 receptor agonists, was associated with an increased risk of lower limb amputation and diabetic ketoacidosis, but not with other serious adverse events of current concern.
Sodium glucose cotransporter 2 (SGLT2) inhibitors present a valuable new therapeutic option for the treatment of type 2 diabetes, but concerns have been raised regarding their safety. In the CANVAS Program, patients randomised to the SGLT2 inhibitor canagliflozin experienced significantly higher rates of lower limb amputation (hazard ratio 1.97, 95% confidence interval 1.41 to 2.75) and bone fracture (1.26, 1.04 to 1.52) compared with patients receiving placebo.1 Case reports to the United States Food and Drug Administration’s (FDA) Adverse Event Reporting System (FAERS) have indicated that SGLT2 inhibitors could cause diabetic ketoacidosis,23 acute kidney injury,45 and serious urinary tract infection.2 SGLT2 inhibitors increase blood viscosity by inducing mild diuresis, which suggests that the risk of venous thromboembolism could be increased.67 The FDA is also investigating reports of acute pancreatitis associated with the use of SGLT2 inhibitors.8910
Completed and ongoing clinical trials are conducted in selected populations and are too small to assess rare adverse events, whereas analyses of case reports have substantial limitations, including a lack of denominators and possible reporting bias. Only four controlled large scale observational studies have been conducted.11121314 Each study investigated one selected adverse event and three studies used study designs with method limitations,111213 including immortal time bias and lag time bias.15 Therefore, the safety of SGLT2 inhibitors is still uncertain.
In this register based cohort study of patients from routine clinical practice, we used nationwide data from Sweden and Denmark to assess whether the use of SGLT2 inhibitors, as compared with an active comparator (glucagon-like peptide 1 (GLP1) receptor agonists), is associated with an increased risk of seven serious adverse events of current concern.
We conducted a cohort study from July 2013 to December 2016, with data from nationwide health and administrative registers in Sweden and Denmark. The registers included population registers (vital status, demographics), patient registers (comorbidities, outcomes), prescription registers (study drugs, comedications), Statistics Denmark and Statistics Sweden (socioeconomic variables), and the Swedish National Diabetes Register (glycated haemoglobin level, blood pressure, albuminuria, estimated glomerular filtration rate, body mass index, and smoking status), which are described in the supplementary materials.
We used an active comparator new-user study design and controlled for a wide range of potential confounders (patient characteristics that might be associated with both the outcome and the decision to initiate a drug) through a non-parsimonious propensity score model to minimise the risk of bias, including confounding by indication.16 We used GLP1 receptor agonists as the active comparator because of important shared features with SGLT2 inhibitors, including use in similar clinical situations (ie, as second line or third line diabetes drugs, with both drug classes likely considered for patients at high cardiovascular risk) and similar temporal trends of use in the two countries,11718192021 but no known associations with the outcomes investigated in this study.
We included all patients in the two countries, aged 35 or over, who filled their first prescription for either an SGLT2 inhibitor or a GLP1 receptor agonist during the study period; the date of filling the first prescription constituted cohort entry. We excluded patients who had previously filled prescriptions for any of the study drugs; had no hospital contact or use of prescription drugs in the previous year; had dialysis or renal transplantation, endstage illness, drug misuse, or severe pancreatic disorders; or had a hospital stay for any reason in the 30 days before cohort entry (supplementary materials, table 1).
Patients were considered exposed to the study drug if prescriptions were refilled before the estimated end date of the most recent prescription. We included a grace period of 90 days to account for irregular drug intake and to capture events that occured shortly after treatment cessation (supplementary materials, table 2).
We estimated propensity scores by using logistic regression for the probability of starting an SGLT2 inhibitor conditional on the status of 66 covariates, defined and selected a priori, including sociodemographic characteristics, comorbidities, comedications, and healthcare utilisation, as current at cohort entry in Sweden and Denmark (supplementary materials, table 3). The nationwide registers provided complete information on disease history (for the previous 10 years) and prescription drug use in each country before cohort entry. Missing data on place of birth (<1%), civil status (<1%), and education (3%) were handled with use of missing categories.22
We matched SGLT2 inhibitor and GLP1 receptor agonist users (1:1 ratio, by country) according to propensity score, by using the nearest neighbour matching algorithm (caliper width 0.2 of the standard deviation of the logit score).2324 Analyses were performed in a pooled dataset of the two countries.
The primary outcome measures were lower limb amputation, bone fracture, diabetic ketoacidosis, acute kidney injury, serious urinary tract infection, venous thromboembolism, and acute pancreatitis, as captured from the patient registers. We also conducted analyses restricted to toe or metatarsal amputation and to major osteoporotic fracture as additional outcomes. Supplementary materials, table 4 shows outcome definitions according to ICD-10 (international classification of diseases, 10th edition) codes and procedure codes. Diagnoses recorded in Scandinavian health registers generally have high sensitivity and positive predictive value,2526 although validation studies have only been conducted for some of the outcomes in the present study. Supplementary materials, table 5 shows validation studies in the national patient registers of diagnostic and procedure codes related to those used in our study.
Patients were followed from cohort entry to treatment cessation, crossover to the other study drug (ie, initiation of GLP1 receptor agonists among SGLT2 inhibitor users and vice versa), the outcome event, death, emigration, or the end of the study period (31 December 2016). We used Cox proportional hazards regression to calculate hazard ratios, analysing each outcome independently. The absolute risk difference was calculated as hazard ratio−1 multiplied by the rate in the comparator group.
We performed subgroup analyses for all outcomes by country, sex, age group, and history of major cardiovascular disease (supplementary materials, table 6). In additional analyses of diabetic ketoacidosis and acute kidney injury, we truncated follow-up at six months as reports to FAERS indicate that these events can occur shortly after treatment initiation.234 For diabetic ketoacidosis and lower limb amputation, we also performed a subgroup analysis according to insulin treatment status at baseline and according to history of peripheral artery disease or lower limb amputation, respectively (supplementary materials, table 7).
For all outcomes, we conducted four sensitivity analyses. Firstly, we used an intention-to-treat exposure definition. In this analysis, patients were considered exposed to the study drug throughout follow-up regardless of treatment cessation or crossover to the other study drug. Secondly, we used an additionally adjusted model. In this analysis, for the propensity score-matched cohort in Sweden, we adjusted the Cox models for additional covariates which provided information about disease severity and comorbidities related to diabetes, including glycated haemoglobin level, blood pressure, albuminuria, estimated glomerular filtration rate, body mass index, and smoking status (supplementary materials, table 8). Given the proportion of missing data for these variables (supplementary materials, table 8), we used multiple imputation (Markov chain Monte Carlo method) to handle missing data27; analyses were conducted by using 10 imputed datasets which were combined. Thirdly, we used a grace period of 30 days in the definition of current use of the study drug. Finally, we adjusted the models for country. We performed additional sensitivity analyses for bone fracture, using an outcome definition only including fractures that led to hospital stays, and excluding users of thiazolidinediones as these drugs have been associated with an increased risk of fracture.28 We performed additional sensitivity analysis for acute pancreatitis, excluding users of dipeptidyl peptidase 4 (DPP4) inhibitors because these drugs have been associated with acute pancreatitis.29 We performed the analysis with SAS version 9.4 software.
No patients were involved in setting the research question or the outcome measures, nor were they involved in developing plans for recruitment, design, or implementation of the study. No patients were asked to advise on interpretation or writing up of results. There are no plans to disseminate the results of the research to study participants or the relevant patient community.
Figure 1 shows that we identified 21 008 new users of SGLT2 inhibitors and 27 278 new users of GLP1 receptor agonists who fulfilled the criteria for study eligibility. Table 1 shows that, compared with GLP1 receptor agonist users, SGLT2 inhibitor users were older, more likely to be men and to use DPP4 inhibitors at baseline, and less likely to have an obesity diagnosis and to use insulin at baseline. Figure 1 and table 1 show that after 1:1 propensity score matching, 34 426 patients (17 213 in each group) were included in the study cohort; the two groups were well balanced on all baseline characteristics. Of the SGLT2 inhibitor users, 61% used dapagliflozin, 38% empagliflozin, and 1% canagliflozin. Patient characteristics before and after propensity score matching for each country separately are shown in supplementary materials, tables 9 and 10. In the analyses of the seven primary outcome measures, median follow-up time in the cohort ranged between 270 and 274 days (supplementary materials, table 11).
Table 2 shows the results of the primary outcome analyses. Use of SGLT2 inhibitors, as compared with GLP1 receptor agonists, was associated with an increased risk of lower limb amputation (hazard ratio 2.32, 95% confidence interval 1.37 to 3.91) and diabetic ketoacidosis (2.14, 1.01 to 4.52) but not with bone fracture (1.11, 0.93 to 1.33), acute kidney injury (0.69, 0.45 to 1.05), serious urinary tract infection (0.89, 0.67 to 1.19), venous thromboembolism (0.99, 0.71to 1.38), or acute pancreatitis (1.16, 0.64 to 2.12).
The hazard ratio for use of SGLT2 inhibitors versus GLP1 receptor agonists was 1.55 (95% confidence interval 0.87 to 2.77) for toe or metatarsal amputation and 1.05 (0.80 to 1.39) for major osteoporotic bone fracture (supplementary materials, table 12).
Subgroup and additional analyses
Figure 2 shows the results of the subgroup analyses. For each outcome, hazard ratios in analyses by country, sex, age group, and history of cardiovascular disease were similar to those observed in the primary analyses, although several of these subgroup analyses were based on a limited number of events. Hazard ratios were also similar in the analysis of diabetic ketoacidosis according to insulin treatment status and in the analysis of lower limb amputation according to history of peripheral artery disease or lower limb amputation. Hazard ratios in the additional analyses of diabetic ketoacidosis and acute kidney injury in which follow-up was truncated at six months after treatment initiation were similar to those in the primary analyses (supplementary materials, table 14).
Figure 3 shows the results of the sensitivity analyses. Hazard ratios were similar to those in the primary analyses when using an exposure definition on intention-to-treat basis, a grace period of 30 days in the definition of current use of the study drug, and when adjusting the models for country. In the Swedish cohort (baseline characteristics in supplementary materials, table 15), additional adjustment of the models for glycated haemoglobin, blood pressure, albuminuria, estimated glomerular filtration rate, body mass index, and smoking status did not materially affect the findings. In the analysis of bone fracture, hazard ratios were not affected by the exclusion of recent users of thiazolidinediones or by the use of an outcome definition that only included fractures leading to hospital stays. In the analysis of acute pancreatitis, hazard ratios were not affected by the exclusion of recent users of DPP4 inhibitors.
Using nationwide registers in two countries, we examined the risk of serious adverse events associated with the use of SGLT2 inhibitors. As compared with GLP1 receptor agonists, SGLT2 inhibitor use was associated with twofold increases in the risk of both lower limb amputation and diabetic ketoacidosis, but no noticeable increase in risk was observed for the other investigated adverse events. The findings should be interpreted in the context of limitations of observational studies and the uncertainty of the effect estimates. Based on the upper limit of the confidence interval, our findings are inconsistent with a relative risk increase of more than 33% for bone fracture, 5% for acute kidney injury, 19% for serious urinary tract infection, 38% for venous thromboembolism, and 112% for acute pancreatitis.
The twofold increase in risk of lower limb amputation associated with the use of SGLT2 inhibitors observed in our study is in line with the findings from the CANVAS Program, which randomised patients with high cardiovascular risk to canagliflozin or placebo.1 Importantly, our results were consistent in patients with and without cardiovascular disease and with and without peripheral arterial disease or previous amputation; the event rates, however, were substantially higher in the subgroups with such history. A pooled analysis of clinical trials of dapagliflozin was underpowered to assess lower limb amputations,30 and no imbalance in this outcome in patients receiving empagliflozin versus placebo was observed in the EMPA-REG OUTCOME trial.31 Whether the increase in lower limb amputations is a class effect for SGLT2 inhibitors, or specific to individual drugs, needs further study.
In our study, the use of SGLT2 inhibitors was not associated with the risk of bone fracture. Increased risk of fractures in patients using canagliflozin was observed in the CANVAS Program but not in the other trials of canagliflozin.132 A meta-analysis of clinical trials revealed no imbalance in fracture rates between those treated with empagliflozin versus placebo.33 Falls owing to volume depletion has been suggested as an underlying mechanism for fractures during use of SGLT2 inhibitors, but it has also been suggested that the drugs affect bone metabolism, as indicated by decreased hip bone mineral density after treatment with canagliflozin.34 However, in analyses restricted to osteoporotic fractures, our results remained similar.
The twofold increased risk of diabetic ketoacidosis observed in our study is in accordance with an analysis of insurance claims in the US, which showed an increase in risk that was of similar magnitude, comparing SGLT2 inhibitors with DPP4 inhibitors.14 Analyses of FEARS data showed higher spontaneous reporting rates of diabetic ketoacidosis in SGLT2 inhibitor users than in DPP4 inhibitor users,3 and rates of this adverse event were numerically higher among those receiving active treatment in the CANVAS Program (0.6 v 0.3 events per 1000 person years),1 and in a meta-analysis of clinical trials of SGLT2 inhibitors (odds ratio 1.96, 95% confidence interval 0.77 to 4.98).35 In addition, higher rates of diabetic ketoacidosis have been observed among those receiving SGLT2 inhibitors in trials that include patients with type 1 diabetes.3637 Viewed together, a relatively coherent picture of an increased risk of diabetic ketoacidosis associated with the use of SGLT2 inhibitors has emerged, although the absolute risk increase seems small.
Reports to FAERS and trials showing a transient decrease in glomerular filtration rate after initiation of SGLT2 inhibitors have given rise to the concern that SGLT2 inhibitors might lead to acute kidney injury.453038 In our study, we found no such association; these findings are in line with those from meta-analyses of trials of SGLT2 inhibitors,303335 the EMPA-REG OUTCOME trial,38 and the CANVAS Program where rates of acute kidney injury were similar or numerically lower among patients randomized to SGLT2 inhibitors.1
We found no association between the use of SGLT2 inhibitors and the risk of serious urinary tract infection. Whereas reports to FAERS and an imbalance in the proportion of patients experiencing urosepsis in the EMPA-REG OUTCOME trial (0.4% empagliflozin v 0.1% placebo) indicated that SGLT2 inhibitors can lead to serious urinary tract infections,218 such associations were not observed in the CANVAS Program and in a meta-analysis of empagliflozin trials.133
We found no increase in the risk of venous thromboembolism associated with the use of SGLT2 inhibitors. It has been speculated that SGLT2 inhibitors might increase the risk of venous thromboembolism by increasing blood viscosity.67 A meta-analysis of clinical trials, including data from regulatory submissions, showed a hazard ratio of 1.54 (95% confidence interval 0.63 to 3.79) for venous thromboembolism in patients receiving active treatment versus placebo,6 but there was no imbalance in event rates in the subsequently presented CANVAS Program or in meta-analyses of SGLT2 inhibitor trials.1303335
Reports to FAERS have prompted the FDA to investigate the risk of acute pancreatitis associated with use of SGLT2 inhibitors.8910 Clinical trials have been too small to assess this potential adverse event. We found no association between the use of SGLT2 inhibitors and acute pancreatitis.
Strengths and weaknesses in relation to other studies
Using nationwide registers to include data from a large number of patients seen in routine clinical practice and implementing a controlled study design that sought to limit the risk of different sources of bias, our study substantially expands on current knowledge regarding the safety of SGLT2 inhibitors and quantifies the risk of serious adverse events potentially linked to the drug class. To date, four controlled large scale observational studies, each investigating one selected outcome (lower limb amputation and diabetic ketoacidosis),11121314 have been conducted on SGLT2 inhibitors and potential adverse events. Three of these studies used a design that permitted patients with a history of comparator drug use to enter the cohort.111213 Therefore, the studies could suffer from compromised confounding control,16 as indicated by the imbalance in diabetic drugs at baseline between users of SGLT2 inhibitors versus comparators, even after propensity score matching. In addition, the implementation of a hierarchical selection of SGLT2 inhibitor treatment episodes in those studies, gives rise to immortal time bias in patients initiating an SGLT2 inhibitor after having used a comparator drug during the study period.11121315 Further, lag time bias could result from the use of any other glucose lowering drug as comparator,15 because some of these drugs are typically used at earlier or later stages of diabetes than SGLT2 inhibitors.19
We attempted to limit the risk of confounding by using a new-user design in which patients had no history of either study drug at cohort entry, a non-parsimonious propensity score and rigorous matching, as well as GLP1 receptor agonists as the comparator. In sensitivity analyses in Sweden (61% of the overall cohort), we also adjusted for glycated haemoglobin level, blood pressure, albuminuria, estimated glomerular filtration rate, body mass index, and smoking status; the consistency of results between these analyses and our main analyses indicate that confounding owing to these variables was minimal, although some of these analyses had few events for each variable used for multivariate adjustment.
Our study has limitations. Firstly, we performed analyses for SGLT2 inhibitors as a drug class; dapagliflozin (61%) and empagliflozin (38%) were the most common SGLT2 inhibitors in our study population and use of canagliflozin was rare. The mechanisms underlying the potential adverse events associated with SGLT2 inhibitors are unknown. Class-wide effects, including volume depletion (lower limb amputation,39 bone fracture,32 diabetic ketoacidosis,40 acute kidney injury,41 and venous thromboembolism),6 increased levels of phosphate and secondary hyperparathyroidism (bone fracture),42 non-insulin dependent glucose lowering and increased glucagon levels (diabetic ketoacidosis),40 uricosuria (acute kidney injury),41 and glucosuria (serious urinary tract infection) have been suggested.43 However, pharmacologic properties and off-target effects specific to individual drugs have been described and reduced hip bone mineral density has been observed after treatment with canagliflozin.344445 Uncertainty regarding the potential effects specific to individual SGLT2 inhibitors remains and assessment of adverse events for individual SGLT2 inhibitors remains a topic for additional examination.
Secondly, the number of events in many of the subgroup analyses were low, resulting in wide confidence intervals. More well powered analyses will require further accumulation of clinical data on the use of SGLT2 inhibitors.
Thirdly, we defined drug exposure based on filled prescriptions; low adherence might bias the results towards the null.
Fourthly, although diagnoses recorded in Scandinavian health registers generally have high sensitivity and positive predictive value,2526 validation studies of the outcomes in our study have only been performed for venous thromboembolism, acute pancreatitis, and certain fractures, infections, and amputations.25264647 Outcome misclassification could have introduced bias in our analyses; such misclassification is unlikely to be differential in the context of this study design and would thus only introduce bias, if any, towards the null. A study from the US showed that diagnostic codes of acute kidney injury in administrative databases have high specificity although registration of events could be incomplete48; the event rates observed in our study should be interpreted with caution.
Finally, residual and unmeasured confounding affecting the findings in our study cannot be ruled out.
In this analysis of nationwide registers from two countries, SGLT2 inhibitors, as compared with GLP1 receptor agonists, were associated with an increased risk of lower limb amputations and diabetic ketoacidosis, but not with other serious adverse events of current concern.
What is already known on this topic
Sodium glucose cotransporter 2 (SGLT2) inhibitors are increasingly popular drugs for the treatment of type 2 diabetes
Data from clinical trials, case reports, and observational studies have indicated that use of this drug class could be associated with serious adverse events, including lower limb amputation, bone fracture, diabetic ketoacidosis, acute kidney injury, serious urinary tract infection, venous thromboembolism, and acute pancreatitis
What this study adds
The use of SGLT2 inhibitors is associated with twofold increases in the risk of lower limb amputation and diabetic ketoacidosis
Contributors: PU, HS, and BP initiated the study and had access to all the data for this project. HS performed the statistical analysis. PU and BP wrote the first draft of the paper. All authors contributed to the acquisition, analysis, or interpretation of data and to the critical revision of the manuscript for important intellectual content. PU, HS, and BP had full access to the data in the study and can take responsibility for the integrity of the data and the accuracy of the data analysis. BP supervised the study and is the guarantor. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted.
Funding: The study was supported by the Swedish Heart-Lung Foundation, the Swedish Cancer Society, the Nordic Cancer Union, the Novo Nordisk Foundation, and the Swedish Society for Medical Research. BP was supported by an investigator grant from the Strategic Research Area Epidemiology program at Karolinska Institutet. HS was supported by a career development investigator grant from Lundbeck Foundation. The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Competing interests: All authors have completed the Unified Competing Interest form at www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and have the following declarations: CJ reports personal fees from Pfizer and Bayer outside the submitted work; BE reports personal fees from Amgen, AstraZeneca, Boerhringer Ingelheim, Eli Lilly, Merck Sharp and Dohme, Mundipharma, Navamedic, Novo Nordisk, and RLS Global outside the submitted work, and grants from Sanofi outside the submitted work; and SG reports personal fees and research grants from AstraZeneca, Boehringer Ingelheim, Eli Lilly, Merck Sharp and Dohme, Novo Nordisk, and Sanofi outside of the submitted work. The other authors did not have any potential conflicts to report.
Ethical approval: The study was approved by the Regional Ethics Committee in Stockholm, Sweden, and the Danish Data Protection Agency. In Denmark, ethics approval is not required for register based research. Informed consent was not required.
Data sharing: No additional data are available.
The lead author (PU) affirms that the manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned have been explained.
This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.