aMultiple reasons for exclusion may apply.
bStopped practicing (eg, retirement).
cNot including outcome data (ie, antibiotic prescriptions).
Circles indicate medians; intervals are 25% and 75% percentiles.
The squares indicate intention-to-treat analysis; the circles indicate on-intervention/per-protocol analysis.
eFigure. Example of Prescription Feedback
eTable 1. Intention To Treat Analysis
eTable 2. On-intervention/Per-Protocol Analysis
eTable 3. Subgroup and Sensitivity Analyses
eAppendix. Search strategy for “Research in Context”
Customize your JAMA Network experience by selecting one or more topics from the list below.
Hemkens LG, Saccilotto R, Reyes SL, et al. Personalized Prescription Feedback Using Routinely Collected Data to Reduce Antibiotic Use in Primary Care: A Randomized Clinical Trial. JAMA Intern Med. 2017;177(2):176–183. doi:https://doi.org/10.1001/jamainternmed.2016.8040
Does quarterly antibiotic prescription feedback to primary care physicians over 2 years reduce antibiotic use when implemented in a complex health care system?
This nationwide pragmatic randomized trial included 2900 Swiss primary care physicians. Physicians receiving feedback prescribed the same amount of antibiotics to all patients as physicians without feedback. Although physicians receiving feedback prescribed fewer antibiotics to younger patients, this finding was not consistent over the entire intervention period.
These findings suggest that quarterly feedback does not change overall antibiotic prescribing. Whether antibiotic use can be reduced in some patient groups remains to be shown.
Feedback interventions using routinely collected health data might reduce antibiotic use nationwide without requiring the substantial resources and structural efforts of other antibiotic stewardship programs.
To determine if quarterly antibiotic prescription feedback over 2 years reduces antibiotic use when implemented in a complex health care system.
Design, Setting, and Participants
Pragmatic randomized trial using routinely collected claims data on 2900 primary care physicians with the highest antibiotic prescription rates in Switzerland.
Physicians were randomized to quarterly updated personalized antibiotic prescription feedback over 2 years (n = 1450) or usual care (n = 1450). Feedback was provided both by mail and online from October 2013 to October 2015 and was supported by an initial 1-time provision of evidence-based guidelines.
Main Outcomes and Measures
The primary outcome was the prescribed defined daily doses (DDD) of any antibiotic to any patient per 100 consultations in the first year analyzed by intention-to-treat. We further analyzed prescriptions of specific antibiotics, age groups, and sex for the first and second year to investigate persistency of effects over time.
The 2900 physicians had 10 660 124 consultations over 2 years of follow-up, prescribed 1 175 780 packages of antibiotics with 10 290 182 DDD. Physicians receiving feedback prescribed the same amount of antibiotics to all patients in the first year (between-group difference, 0.81%; 95% CI, −2.56% to 4.30%; P = .64) and second year (between-group difference, −1.73%; 95% CI, −5.07% to 1.72%; P = .32) compared with the control group. Prescribing to children aged 6 to 18 years was −8.61% lower in the feedback than in the control group in the first year (95% CI, −14.87% to −1.90%; P = .01). This difference diminished in the second year (between-group difference, −4.10%; 95% CI, −10.78% to 3.07%; P = .25). Physicians receiving feedback prescribed fewer antibiotics to adults aged 19 to 65 years in the second year (between-group difference, −4.59%; 95% CI, −7.91% to −1.16%; P < .01). Prescribing to other patient groups or of specific antibiotic types was not significantly different between groups.
Conclusions and Relevance
This nationwide antibiotic stewardship program with routine feedback on antibiotic prescribing was not associated with a change of antibiotic use. In older children, adolescents, and younger adults less antibiotics were prescribed, but not consistently over the entire intervention period.
clinicaltrials.gov Identifier: NCT01773824
Antibiotic resistance is closely correlated with antibiotic use in primary care and is a major threat to public health.1 Various antibiotic stewardship approaches involving communication training, specific education interventions, point-of-care testing, electronic decision support systems, and delayed prescribing have the strongest evidence base for beneficial effects on antibiotic prescribing.2,3 However, their system-wide implementation would often require time, major resources, and substantial structural efforts. Prescription feedback interventions are less resource intensive, timely implementable and beneficial effects have been found across various medical fields.4 For antibiotic prescribing, however, current evidence is inconsistent and scarce,2,3 in particular for a large-scale system-wide implementation with no direct patient or physician contact.5-13 Existing large-scale trials evaluating prescription feedback did not continuously or specifically target high prescribers, were of short duration, and were implemented in settings not sharing the complexity of the Swiss or US health care system.5,6
We investigated in a nationwide trial the feasibility and effectiveness of a large-scale, quarterly prescription feedback intervention on antibiotic use in primary care over 2 years using routinely collected claims data in Switzerland.
We did a pragmatic nationwide randomized parallel group trial. We identified the primary care physicians with the highest antibiotic prescription rates in Switzerland using routinely collected claims data of prescriptions of antibiotics and outpatient consultations collected by SASIS, a data warehouse company of an umbrella organization of Swiss statutory health insurers (Santésuisse). These data are collected by over 60 statutory health insurers covering 64% of the Swiss population (5.1 million residents).14 We included among all board certified primary care physicians the 2900 top antibiotic prescribers (based on prescribed defined daily doses [DDD] per 100 consultations in the year prior to randomization; 2900 was the calculated necessary sample size; 2484 [86%] prescribed above the median). Physicians were identified via the Swiss central physician registry that allows health insurers to identify all physicians and claims for any of their services or prescriptions. Registration numbers belonging to any ambulatory facilities of hospitals were excluded. Physicians with fewer than 100 patients per year were excluded. We used no other selection criteria. No informed consent was obtained but all physicians in the intervention group could decline to be contacted at any time. This trial was approved by all ethic committees responsible for all 26 cantons of Switzerland. Details on the rationale and design are described elsewhere.14
Eligible physicians were 1:1 randomized to the intervention or control group by an independent researcher, using a computer algorithm. Physicians in the intervention group were not aware of being part of a controlled trial; physicians in the control group were not informed. Investigators were blinded owing to the anonymized nature of the trial. The outcome assessment was, formally, blinded because all study-relevant data were collected by health insurance personnel not involved in the study.
Physicians in the intervention group received a quarterly updated personalized prescription feedback. Physicians in the control group received no material.
In quarterly intervals, beginning in October 2013, we sent physicians in the intervention group a letter enclosing a quarterly updated single-page graphical overview showing the individual amount of antibiotic prescriptions per 100 consultations in the preceding months and displaying the adjusted average in peer physicians, that is the entire population of Swiss primary care physicians (eFigure in the Supplement). This letter also included an individual access code to the study website, where we offered more detailed online prescription feedback (for example with details on the prescriptions per age group or sex or for certain antibiotic types) and answers to frequently asked questions on antibiotic use. Overall, 8 postal feedbacks were provided, the last in July 2015.
With the first mailing, we sent further information providing basic details on the study and its rationale, clarifying privacy and data protection issues, and justifying the use of anonymized insurance data for research purposes. The intervention was described as quality improvement program and no details about the conduct and design of the trial were provided. This mailing also included a response postcard for opting-out of the intervention and evidence-based guidelines for optimized antibiotic use in primary care, adapted for Switzerland. The guidelines focused on the 7 most frequent reasons for antibiotic prescribing in primary care (acute unspecific upper respiratory tract infection, sore throat/acute tonsillitis/pharyngitis, acute rhinosinusitis, acute otitis media, acute bronchitis, community-acquired pneumonia, and uncomplicated urinary tract infection). There was no other change of concomitant care or practice.
The primary outcome was the prescribed DDD of any type of antibiotics to any patient per 100 consultations (DDD/100c) in the first year. The prescribed DDD is “the assumed average maintenance dose per day for a drug used for its main indication in adults.”15 Antibiotic prescriptions (in DDD/100c) were further assessed in young children (0 to 5 years), older children and adolescents (6 to 18 years), younger adults (19 to 65 years), elderly (older than 65 years), and women or men. We also assessed prescriptions by specific antibiotic types (ie, tetracyclines, amphenicols, β-lactams/penicillins, other β-lactams, sulfonamides/trimethoprim, macrolides/lincosamides/streptogramins, aminoglycosides, quinolones, other antibacterials, unspecified/unknown). We measured the outcomes for the first and second year separately to evaluate early on intervention effects in the first and any weaning effects in the second year.
We calculated a necessary sample size of 2900 physicians to detect with a power of 90% (1-β) and a 2-sided significance level of .05 (α) a 5% between-group difference for the primary outcome in the first year. We assumed that physicians who regularly access the online service (expected to be 30% of the intervention group physicians) would reduce their antibiotic prescriptions by 5%, the remaining physicians by 2% with no change in the control group.
We log-transformed the data for analysis because antibiotic prescription data were asymmetrical. We obtained least square means from the linear mixed model and back-transformed them to estimate and report medians in the original scale. We estimated the intervention effect separately for each year by comparing the between-group (intervention minus control) mean difference of log-transformed antibiotic prescription rates in the year before randomization with the between-group difference in the first and second year after randomization using a linear mixed model. The model included as fixed effects the randomized group (intervention or control); time (baseline period, first year, second year of follow-up); an interaction of randomized group with time; and predefined baseline covariates. As random effect, the physician identifier was included.
All baseline covariates were selected prior to unblinding intervention allocation. They were: the total number of consultations at baseline (low, medium, high, according to tertile); medication dispensing status of the physician (self-dispensing, not self-dispensing, or mixture of both); and patient-mix type treated by the physician. The patient-mix was derived by a hierarchical cluster analysis of prescriptions related to 4 major comorbidities, ie, immunosuppression (including rheumatic disease, cancer, HIV/AIDS, organ transplant), asthma/COPD, cardiovascular diseases, and diabetes. We used the Ward method to define 2 clusters of physicians who have either a patient-mix type including many or including few patients with these comorbidities.16 For very few physician identifiers (n = 28), the routine data showed implausibly high values for baseline covariates, or an atypical combination of values for baseline variables indicating group practices (eg, extremely high use of antiretroviral drugs) or data errors. We thus excluded them from the cluster analysis using the Mahalanobis distance (highest 1% values).17
The main analysis is by intention-to-treat not excluding physicians who declined to receive feedback. When we had no prescribing information for a certain month we assumed that a physician actually prescribed no antibiotics whenever there was at least 1 consultation in this month, otherwise we kept the value of antibiotics as missing. We did not impute any missing data. In some cases there was no information on baseline characteristics used for the linear model; then we excluded those physicians from this analysis.
We conducted a combined on-intervention/per-protocol analysis excluding intervention group physicians who opted out (and thus did not receive the intervention over 24 months) and excluding physicians who, in retrospect, were ineligible based on our inclusion criteria when using the most recent and complete data set to verify the selection criteria (see below). This analysis was explorative, fully acknowledging the associated selection biases.
We conducted several sensitivity analyses. We used a different outcome definition (prescribed packages instead of DDD), we excluded physicians who opted out, we calculated a model without using the baseline covariates, we excluded physicians with incomplete follow-up data, and we excluded physicians with outlying outcome values at any time point. We also used the Wilcoxon rank test instead of a linear mixed model.
Differences between self-dispensing and non–self-dispensing physicians and between physicians treating many or few patients with multiple comorbidities were explored in prespecified subgroup analyses. We tested the effect modification with a formal test for interaction in the linear model. All analyses were done using SAS statistical software (version 9.4, SAS Institute Inc) and R statistical software (version 3.3.0., the R Foundation).
The data used for recruitment, intervention, and main outcome measurement was routinely collected by statutory health insurers for claims of drug prescriptions and health care services. Owing to administrative processes, there is a lag between prescription and reimbursement dates. After 3 months, 84% of prescriptions and health care services are recorded in the claims database, after 6 months approximately 97%. For each quarterly feedback we used an updated data set, the final analysis data set was provided in December 2015. Further details are elsewhere.14
In the first study year, the feedback on antibiotic prescriptions per month was based on the date of the claim (eFigure, A, in the Supplement). This was changed in the second year to the actual prescription dates when the provided data for year 2 allowed this (eFigure, B, in the Supplement). All analyses are based on the prescription dates.
Of 2900 randomized physicians, all 1450 physicians in the intervention group received the evidence-based guidelines and first feedback information (Figure 1). Of the 1450 physicians, 211 (14.6%) opted out later. We used data from 2814 physicians for the intention-to-treat analysis, including the 211 physicians who opted out. Baseline characteristics were similar between the intervention and control group (Table).
Across all physicians there were 10 660 124 consultations over the 2 years of follow-up with 1 175 780 packages of antibiotics prescribed, corresponding to 10 290 182 daily doses. Antibiotic prescription rates were higher in the baseline period (median 100.7 DDD/100c) than in the first and second year of follow-up in both the intervention and the control group (first year 90.4 DDD/100c, second year 92.1 DDD/100c) with strong seasonal variation (Figure 2).
There was no difference in antibiotic prescribing to all patients between physicians receiving the intervention and physicians in the control group in the first and in the second year (between group difference, 0.81%; 95%CI, −2.56% to 4.30% and between group difference, −1.73%; 95%CI, −5.07% to 1.72%) (Figure 3 and Figure 4) (eTable 1 in the Supplement). The intervention was associated with reduced prescribing to older children and adolescents (6 to 18 years) in the first year (between group difference, −8.61%; 95% CI, −14.87% to −1.90%). This association diminished in the second year (between group difference, −4.10%; 95% CI, −10.78% to 3.07%; P = .20 for year vs year effect). Physicians in the intervention group prescribed less antibiotics to adults aged 19 to 65 years only in the second year (between group difference, −4.59%; 95% CI, −7.91% to −1.16%; P = .01 for year vs year effect) compared with the control group. We found no between-group differences in prescribing to other patient groups or in prescribing of specific types of antibiotics, only for macrolides there was a statistically significant difference in the second year with fewer prescriptions in the intervention group (between group difference, −5.71%; 95% CI, −10.75% to −0.38%).
The explorative on-intervention/per-protocol analysis showed lower overall prescribing (between group difference, −4.06%; 95% CI, −7.53% to −0.45%) (Figure 3 and Figure 4) (eTable 2 in the Supplement) and lower prescribing to elderly patients in the second year by physicians receiving feedback (between group difference, −4.26%; 95% CI, −8.33 to −0.01%). All other findings were similar to the main intention-to-treat analysis.
Prespecified subgroup analyses showed no significant differences between self-dispensing and non–self-dispensing physicians (P=.90 for interaction) (eTable 3 in the Supplement) and between physicians treating many or few patients with many comorbidities in both follow-up years (P = .90). All sensitivity analyses supported the main findings showing no differences between the intervention and control group (eTable 3 in the Supplement).
During the entire follow-up, 11% of physicians in the intervention group accessed the online tool.
In this pragmatic nationwide study we randomized 2900 physicians who represent the top antibiotic prescribers in Swiss primary care. We used claims data of more than 5 million Swiss residents. Quarterly personalized prescription feedback over 2 years combined with a 1-time provision of evidence-based guidelines did not affect overall antibiotic prescribing. However, it reduced prescribing to prespecified subpopulations of children and adolescents by −8.6% in the first year and to young and middle-aged adults in the second year of the intervention by −4.6%. To illustrate the potential public health impact of a system-wide prescription feedback to 2900 top prescribers, given yearly consultations in the range of 317 000 and 3 336 000 in these age groups, we would estimate that such an intervention would lead to an overall reduction of 29 500 daily doses prescribed to children and adolescents and 165 000 doses prescribed less to young and middle-aged adults. However, the findings for these populations were not consistent over the intervention period and need to be cautiously interpreted.
The costs for the intervention were low despite the large scale: the budget for this study, including the design, implementation, and evaluation of the feedback intervention, was in a range of $300 000 but may be higher in other settings.
We observed no impact on prescribing of specific antibiotic types, with the exception of reduced prescribing of macrolides. However, this might be just a chance finding and should be very cautiously interpreted.
We conducted a rapid systematic review to put the results in context, specifically focusing on large, countrywide approaches not involving elements that would be difficult to be implemented on a large scale (such as on-site visits18 or educational elements) (eAppendix in the Supplement). Two trials evaluated feedback on a large scale, involving more than 1000 physicans.5,6 One trial found no impact on antibiotic prescriptions when 2 mailed feedback packets (that addressed also 4 other drug groups besides antibiotics) where given to unselected Australian general practitioners.6 The other trial found that a single feedback letter sent to the top 20% of antibiotic prescribing general practitioners in 2014 in England reduced antibiotic prescribing by 3.3% over 6 months.5 Smaller studies similarly showed inconsistent findings.7-13
In contrast to many previous feedback approaches to reduce outpatient antibiotic prescribing, our intervention used several elements of feedback interventions that are increasingly recognized to enhance feedback effectiveness.19 We targeted individuals with suboptimal baseline performance, aimed to decrease (instead of promote) a behavior, and gave repetitive feedback over a longer period. We gave physicians not only the opportunity to compare their own prescribing behavior to that of peers; they could also evaluate long-term trends in their own prescribing over up to 2 years. We also delivered our feedback from the perspective of clinical colleagues. However, owing to the anonymization, the mailings were sent from a company (SASIS)14 owned by health insurers which might not have been well received by some general practitioners owing to its relation to the payer side and involvement in cost-control activities in the statutory health system. We therefore intended to provide very objective feedback without directive or emotional elements. This rather official character of the sender, however, may have increased the awareness of the feedback. The abovementioned recent trial from England provided feedback that came from England's Chief Medical Officer. However, in the decentralized and federal Swiss health care system there is no official national health authority that we could have involved.
Our study shows that quarterly provided prescription feedback over 2 years is possible at low costs on a nationwide scale. Remarkably, the intervention was feasible in a country with multiple official languages, a complex health care system with numerous health insurers, a large private health care sector, and a highly federal structure allowing for self-dispensing in some cantons. However, the online feedback system met very little response despite promoting the online service with each mailed feedback and repeatedly providing the website log-in information.
We did not disclose the complete study design to participating physicians before completion to preserve generalizability of findings and avoid bias that we expected when physicians would be aware of the trial setting. The feedback was also linked to the individual physician’s registration number and less than 5% of the control group physicians had the same contact data as the intervention group physicians (what might indicate group practices), thus making relevant contamination to the control group unlikely.20 This approach maximized external validity and the routine data collection gave us additionally the unique opportunity to monitor the prescription behavior of those 15% of physicians who refused to receive feedback. When we excluded declining physicians in the on-intervention analysis, we actually observed a stronger (and statistically significant) overall intervention effect compared with the main intention-to-treat analysis. This finding is obviously biased owing to the selective exclusion of the nonparticipating physicians in the intervention group. We could not identify any characteristic of this subgroup beyond the fact that they opted out and saw slightly fewer patients. Thus, physicians who are not approachable with such a kind of intervention seem to prescribe differently for reasons that are not reflected by the routine data. From a public health perspective, the findings from the on-intervention analysis are also informative because they demonstrate that general practitioners who decline routine prescription feedback represent an important special target group for antibiotic stewardship initiatives.
Several issues may explain why we found no association of the feedback with overall reduced prescribing rates.
First, Switzerland has the lowest antibiotic prescription rates in Europe, thus any strategy to decrease antibiotic use may here be more challenging than elsewhere.1,21,22
Second, the use of routinely collected data in its presently available structure introduced some limitations. We had no individual patient data, only data aggregated by month and physician, and we had no data on diagnoses, hospitalizations, or mortality because such data are not routinely provided by Swiss health care providers to health insurers. This did not allow us to determine how many DDD were given to a specific number of patients or to measure patient-relevant benefits and harms. However, we used data on prescriptions for other drugs to estimate the patient mix treated by the physicians. The aggregated nature of the data also precluded more refined time-trend analyses using more detailed patient-level information and we could not provide more tailored or detailed feedback (eg, on guideline concordant disease-specific prescribing, on treatment duration, or comparisons with specific, highly matched peers with almost identical patient populations). Although the database is the largest available, it does not cover the entire Swiss population. Therefore, the feedback could be based only on a part of a physician’s patient population (this was clearly communicated to the physicians).
Third, the feedback was based in the first year on the date of the claim which may have made the interpretation less intuitive. We were able to use the actual antibiotic prescription date in the second year of the intervention. We then juxtaposed the monthly personal prescriptions with that of the peers, but also with the personal previous year prescriptions as a second benchmark. This may explain some of the differences between the first and second year findings.
Fourth, the time lag between consultation or prescription and database entry decreased the directness of the given feedback in the first year and introduced some noise in the data. This noise decreased with continuous data updates but led to the inclusion of some physicians who would have been ineligible based on the most updated data set available 2 years later, however the inclusion of these physicians had no impact on the interpretation of findings, as the sensitivity analyses show. These deficiencies of routine data collection may have led to a smaller intervention effect, but they are probably inevitable in many health care settings with current data infrastructures.
Fifth, the peer comparison may have had negative effects for those receiving positive feedback showing that they are doing better than the average because they might sit back and tend to underachieve. However, this could also be a positive reinforcement to sustain the behavior, which would agree with successful reductions of antibiotic prescriptions in a recent trial that specifically involved such reinforcing feedback.13
Finally, the aggregated routinely collected claims data do not allow drawing any conclusion on patient-relevant benefits and antibiotic resistance in the community.
It remains unclear why the intervention affected prescriptions for some but not all patient populations. This might be just a chance finding owing to numerous subanalyses. Another explanation might be that indications to prescribe antibiotics in older children and younger adults might be perceived as less compelling, and thus allowing more restraint than in elderly patients or very young children where medical reasons (eg, comorbidities) and external pressure (eg, parents’ expectations) might be more compelling.
Quarterly personalized prescription feedback over 2 years combined with a 1-time provision of evidence-based guidelines does not reduce antibiotic use. Whether antibiotic use can be reduced in some patient groups like the younger remains to be shown. Given the low costs for implementation, more intense and better tailored prescription feedback approaches merit further evaluation and it should be shown whether they are associated with patient-relevant benefits and directly impact antibiotic resistance.
Corresponding Author: Heiner C. Bucher, MD, MPH, Basel Institute for Clinical Epidemiology and Biostatistics, University Hospital Basel, CH-4031 Basel, Switzerland (email@example.com).
Published Online: December 27, 2016. doi:10.1001/jamainternmed.2016.8040
Author Contributions: Dr Bucher had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Drs Hemkens and Saccilotto contributed equally to this work.
Concept and design: Hemkens, Saccilotto, Raatz, Widmer, Zeller, Bucher.
Acquisition, analysis, or interpretation of data: Hemkens, Saccilotto, Leon Reyes, Glinz, Zumbrunn, Grolimund, Gloy, Bucher.
Drafting of the manuscript: Hemkens, Widmer, Zeller, Bucher.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Leon Reyes, Zumbrunn.
Obtained funding: Saccilotto, Bucher.
Administrative, technical, or material support: Hemkens, Saccilotto, Zumbrunn, Grolimund, Gloy, Widmer, Bucher.
Supervision: Hemkens, Widmer, Zeller, Bucher.
Conflict of Interest Disclosures: The Basel Institute for Clinical Epidemiology and Biostatistics, University Hospital Basel, Basel, Switzerland, was in 2013 supported by Santésuisse, an umbrella association of Swiss social health insurers. Selene Leon Reyes was consultant for Novartis Pharmaceuticals Corporation in 2014. Oliver Grolimund is employee of SASIS AG, the company providing the routinely collected health data for this study. No other disclosures are reported.
Funding/Support: This study was funded by a grant from the Swiss National Science Foundation (32003B_140997/1) and a grant from the Swiss Academy of Medical Science (Versorgungsforschung Gottfried und Julia Bangerter-Rhyner-Stiftung). The Basel Institute for Clinical Epidemiology and Biostatistics was supported in 2013 by an unrestricted grant from Santésuisse, an umbrella association of Swiss social health insurers.
Role of the Funder/Sponsor: The funders 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.
Additional Contributions: We thank Gerd Laifer, MD, Praxis Hammer, Basel, Switzerland; David Nadal, MD, Division of Infectious Diseases and Hospital Epidemiology, University Children's Hospital of Zurich, Zurich, Switzerland; Christoph Rudin, MD, University Children's Hospital Basel, Basel, Switzerland; Maja Weisser, MD, Division of Infectious Diseases and Hospital Epidemiology, University Hospital Basel, University of Basel, Switzerland; and Antje Welge-Lüssen, MD, Department of Otorhinolaryngology, University Hospital Basel, University of Basel, Switzerland, for reviewing the treatment guidelines. They were not compensated for their contributions. We also thank Stefanie Merlo, Sandra Manz, and Kübra Özoglu, Basel Institute for Clinical Epidemiology and Biostatistics, University Hospital Basel, University of Basel, Switzerland, for administrative assistance.
Create a personal account or sign in to: