Antimicrobial Resistance as Risk Factor for Recurrent Bacteremia after Staphylococcus aureus, Escherichia coli, or Klebsiella spp. Community-Onset Bacteremia.

We investigated links between antimicrobial resistance in community-onset bacteremia and 1-year bacteremia recurrence by using the clinical data warehouse of Europe's largest university hospital group in France. We included adult patients hospitalized with an incident community-onset Staphylococcus aureus, Escherichia coli, or Klebsiella spp. bacteremia during 2017-2019. We assessed risk factors of 1-year recurrence using Fine-Gray regression models. Of the 3,617 patients included, 291 (8.0%) had >1 recurrence episode. Third-generation cephalosporin (3GC)-resistance was significantly associated with increased recurrence risk after incident Klebsiella spp. (hazard ratio 3.91 [95% CI 2.32-6.59]) or E. coli (hazard ratio 2.35 [95% CI 1.50-3.68]) bacteremia. Methicillin resistance in S. aureus bacteremia had no effect on recurrence risk. Although several underlying conditions and infection sources increased recurrence risk, 3GC-resistant Klebsiella spp. was associated with the greatest increase. These results demonstrate a new facet to illness induced by 3GC-resistant Klebsiella spp. and E. coli in the community setting.

Exploring factors shaping antibiotic resistance patterns in Streptococcus pneumoniae during the 2020 COVID-19 pandemic.

Non-pharmaceutical interventions implemented to block SARS-CoV-2 transmission in early 2020 led to global reductions in the incidence of invasive pneumococcal disease (IPD). By contrast, most European countries reported an increase in antibiotic resistance among invasive Streptococcus pneumoniae isolates from 2019 to 2020, while an increasing number of studies reported stable pneumococcal carriage prevalence over the same period. To disentangle the impacts of the COVID-19 pandemic on pneumococcal epidemiology in the community setting, we propose a mathematical model formalizing simultaneous transmission of SARS-CoV-2 and antibiotic-sensitive and -resistant strains of S. pneumoniae. To test hypotheses underlying these trends five mechanisms were built into the model and examined: (1) a population-wide reduction of antibiotic prescriptions in the community, (2) lockdown effect on pneumococcal transmission, (3) a reduced risk of developing an IPD due to the absence of common respiratory viruses, (4) community azithromycin use in COVID-19 infected individuals, (5) and a longer carriage duration of antibiotic-resistant pneumococcal strains. Among 31 possible pandemic scenarios involving mechanisms individually or in combination, model simulations surprisingly identified only two scenarios that reproduced the reported trends in the general population. They included factors (1), (3), and (4). These scenarios replicated a nearly 50% reduction in annual IPD, and an increase in antibiotic resistance from 20% to 22%, all while maintaining a relatively stable pneumococcal carriage. Exploring further, higher SARS-CoV-2 R0 values and synergistic within-host virus-bacteria interaction mechanisms could have additionally contributed to the observed antibiotic resistance increase. Our work demonstrates the utility of the mathematical modeling approach in unraveling the complex effects of the COVID-19 pandemic responses on AMR dynamics.

Collateral impacts of pandemic COVID-19 drive the nosocomial spread of antibiotic resistance: A modelling study.

BACKGROUND: Circulation of multidrug-resistant bacteria (MRB) in healthcare facilities is a major public health problem. These settings have been greatly impacted by the Coronavirus Disease 2019 (COVID-19) pandemic, notably due to surges in COVID-19 caseloads and the implementation of infection control measures. We sought to evaluate how such collateral impacts of COVID-19 impacted the nosocomial spread of MRB in an early pandemic context. METHODS AND FINDINGS: We developed a mathematical model in which Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) and MRB cocirculate among patients and staff in a theoretical hospital population. Responses to COVID-19 were captured mechanistically via a range of parameters that reflect impacts of SARS-CoV-2 outbreaks on factors relevant for pathogen transmission. COVID-19 responses include both "policy responses" willingly enacted to limit SARS-CoV-2 transmission (e.g., universal masking, patient lockdown, and reinforced hand hygiene) and "caseload responses" unwillingly resulting from surges in COVID-19 caseloads (e.g., abandonment of antibiotic stewardship, disorganization of infection control programmes, and extended length of stay for COVID-19 patients). We conducted 2 main sets of model simulations, in which we quantified impacts of SARS-CoV-2 outbreaks on MRB colonization incidence and antibiotic resistance rates (the share of colonization due to antibiotic-resistant versus antibiotic-sensitive strains). The first set of simulations represents diverse MRB and nosocomial environments, accounting for high levels of heterogeneity across bacterial parameters (e.g., rates of transmission, antibiotic sensitivity, and colonization prevalence among newly admitted patients) and hospital parameters (e.g., rates of interindividual contact, antibiotic exposure, and patient admission/discharge). On average, COVID-19 control policies coincided with MRB prevention, including 28.2% [95% uncertainty interval: 2.5%, 60.2%] fewer incident cases of patient MRB colonization. Conversely, surges in COVID-19 caseloads favoured MRB transmission, resulting in a 13.8% [-3.5%, 77.0%] increase in colonization incidence and a 10.4% [0.2%, 46.9%] increase in antibiotic resistance rates in the absence of concomitant COVID-19 control policies. When COVID-19 policy responses and caseload responses were combined, MRB colonization incidence decreased by 24.2% [-7.8%, 59.3%], while resistance rates increased by 2.9% [-5.4%, 23.2%]. Impacts of COVID-19 responses varied across patients and staff and their respective routes of pathogen acquisition. The second set of simulations was tailored to specific hospital wards and nosocomial bacteria (methicillin-resistant Staphylococcus aureus, extended-spectrum beta-lactamase producing Escherichia coli). Consequences of nosocomial SARS-CoV-2 outbreaks were found to be highly context specific, with impacts depending on the specific ward and bacteria evaluated. In particular, SARS-CoV-2 outbreaks significantly impacted patient MRB colonization only in settings with high underlying risk of bacterial transmission. Yet across settings and species, antibiotic resistance burden was reduced in facilities with timelier implementation of effective COVID-19 control policies. CONCLUSIONS: Our model suggests that surges in nosocomial SARS-CoV-2 transmission generate selection for the spread of antibiotic-resistant bacteria. Timely implementation of efficient COVID-19 control measures thus has 2-fold benefits, preventing the transmission of both SARS-CoV-2 and MRB, and highlighting antibiotic resistance control as a collateral benefit of pandemic preparedness.

Sick leave due to COVID-19 during the first pandemic wave in France, 2020.

OBJECTIVES: To quantify the burden of COVID-19-related sick leave during the first pandemic wave in France, accounting for sick leaves due to symptomatic COVID-19 ('symptomatic sick leaves') and those due to close contact with COVID-19 cases ('contact sick leaves'). METHODS: We combined data from a national demographic database, an occupational health survey, a social behaviour survey and a dynamic SARS-CoV-2 transmission model. Sick leave incidence from 1 March 2020 to 31 May 2020 was estimated by summing daily probabilities of symptomatic and contact sick leaves, stratified by age and administrative region. RESULTS: There were an estimated 1.70M COVID-19-related sick leaves among France's 40M working-age adults during the first pandemic wave, including 0.42M due to COVID-19 symptoms and 1.28M due to COVID-19 contacts. There was great geographical variation, with peak daily sick leave incidence ranging from 230 in Corse (Corsica) to 33 000 in Île-de-France (the greater Paris region), and greatest overall burden in regions of north-eastern France. Regional sick leave burden was generally proportional to local COVID-19 prevalence, but age-adjusted employment rates and contact behaviours also contributed. For instance, 37% of symptomatic infections occurred in Île-de-France, but 45% of sick leaves. Middle-aged workers bore disproportionately high sick leave burden, owing predominantly to greater incidence of contact sick leaves. CONCLUSIONS: France was heavily impacted by sick leave during the first pandemic wave, with COVID-19 contacts accounting for approximately three-quarters of COVID-19-related sick leaves. In the absence of representative sick leave registry data, local demography, employment patterns, epidemiological trends and contact behaviours can be synthesised to quantify sick leave burden and, in turn, predict economic consequences of infectious disease epidemics.

Impact of non-pharmaceutical interventions on SARS-CoV-2 outbreaks in English care homes: a modelling study.

BACKGROUND: COVID-19 outbreaks still occur in English care homes despite the interventions in place. METHODS: We developed a stochastic compartmental model to simulate the spread of SARS-CoV-2 within an English care home. We quantified the outbreak risk with baseline non-pharmaceutical interventions (NPIs) already in place, the role of community prevalence in driving outbreaks, and the relative contribution of all importation routes into a fully susceptible care home. We also considered the potential impact of additional control measures in care homes with and without immunity, namely: increasing staff and resident testing frequency, using lateral flow antigen testing (LFD) tests instead of polymerase chain reaction (PCR), enhancing infection prevention and control (IPC), increasing the proportion of residents isolated, shortening the delay to isolation, improving the effectiveness of isolation, restricting visitors and limiting staff to working in one care home. We additionally present a Shiny application for users to apply this model to their facility of interest, specifying care home, outbreak and intervention characteristics. RESULTS: The model suggests that importation of SARS-CoV-2 by staff, from the community, is the main driver of outbreaks, that importation by visitors or from hospitals is rare, and that the past testing strategy (monthly testing of residents and daily testing of staff by PCR) likely provides negligible benefit in preventing outbreaks. Daily staff testing by LFD was 39% (95% 18-55%) effective in preventing outbreaks at 30 days compared to no testing. CONCLUSIONS: Increasing the frequency of testing in staff and enhancing IPC are important to preventing importations to the care home. Further work is needed to understand the impact of vaccination in this population, which is likely to be very effective in preventing outbreaks.

Rapid antigen testing as a reactive response to surges in nosocomial SARS-CoV-2 outbreak risk.

Healthcare facilities are vulnerable to SARS-CoV-2 introductions and subsequent nosocomial outbreaks. Antigen rapid diagnostic testing (Ag-RDT) is widely used for population screening, but its health and economic benefits as a reactive response to local surges in outbreak risk are unclear. We simulate SARS-CoV-2 transmission in a long-term care hospital with varying COVID-19 containment measures in place (social distancing, face masks, vaccination). Across scenarios, nosocomial incidence is reduced by up to 40-47% (range of means) with routine symptomatic RT-PCR testing, 59-63% with the addition of a timely round of Ag-RDT screening, and 69-75% with well-timed two-round screening. For the latter, a delay of 4-5 days between the two screening rounds is optimal for transmission prevention. Screening efficacy varies depending on test sensitivity, test type, subpopulations targeted, and community incidence. Efficiency, however, varies primarily depending on underlying outbreak risk, with health-economic benefits scaling by orders of magnitude depending on the COVID-19 containment measures in place.

Monitoring sick leave data for early detection of influenza outbreaks.

BACKGROUND: Workplace absenteeism increases significantly during influenza epidemics. Sick leave records may facilitate more timely detection of influenza outbreaks, as trends in increased sick leave may precede alerts issued by sentinel surveillance systems by days or weeks. Sick leave data have not been comprehensively evaluated in comparison to traditional surveillance methods. The aim of this paper is to study the performance and the feasibility of using a detection system based on sick leave data to detect influenza outbreaks. METHODS: Sick leave records were extracted from private French health insurance data, covering on average 209,932 companies per year across a wide range of sizes and sectors. We used linear regression to estimate the weekly number of new sick leave spells between 2016 and 2017 in 12 French regions, adjusting for trend, seasonality and worker leaves on historical data from 2010 to 2015. Outbreaks were detected using a 95%-prediction interval. This method was compared to results from the French Sentinelles network, a gold-standard primary care surveillance system currently in place. RESULTS: Using sick leave data, we detected 92% of reported influenza outbreaks between 2016 and 2017, on average 5.88 weeks prior to outbreak peaks. Compared to the existing Sentinelles model, our method had high sensitivity (89%) and positive predictive value (86%), and detected outbreaks on average 2.5 weeks earlier. CONCLUSION: Sick leave surveillance could be a sensitive, specific and timely tool for detection of influenza outbreaks.

A Conceptual Discussion About the Basic Reproduction Number of Severe Acute Respiratory Syndrome Coronavirus 2 in Healthcare Settings.

To date, no specific estimate of R0 for SARS-CoV-2 is available for healthcare settings. Using interindividual contact data, we highlight that R0 estimates from the community cannot translate directly to healthcare settings, with pre-pandemic R0 values ranging 1.3-7.7 in 3 illustrative healthcare institutions. This has implications for nosocomial COVID-19 control.

Optimizing COVID-19 surveillance in long-term care facilities: a modelling study.

BACKGROUND: Long-term care facilities (LTCFs) are vulnerable to outbreaks of coronavirus disease 2019 (COVID-19). Timely epidemiological surveillance is essential for outbreak response, but is complicated by a high proportion of silent (non-symptomatic) infections and limited testing resources. METHODS: We used a stochastic, individual-based model to simulate transmission of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) along detailed inter-individual contact networks describing patient-staff interactions in a real LTCF setting. We simulated distribution of nasopharyngeal swabs and reverse transcriptase polymerase chain reaction (RT-PCR) tests using clinical and demographic indications and evaluated the efficacy and resource-efficiency of a range of surveillance strategies, including group testing (sample pooling) and testing cascades, which couple (i) testing for multiple indications (symptoms, admission) with (ii) random daily testing. RESULTS: In the baseline scenario, randomly introducing a silent SARS-CoV-2 infection into a 170-bed LTCF led to large outbreaks, with a cumulative 86 (95% uncertainty interval 6-224) infections after 3 weeks of unmitigated transmission. Efficacy of symptom-based screening was limited by lags to symptom onset and silent asymptomatic and pre-symptomatic transmission. Across scenarios, testing upon admission detected just 34-66% of patients infected upon LTCF entry, and also missed potential introductions from staff. Random daily testing was more effective when targeting patients than staff, but was overall an inefficient use of limited resources. At high testing capacity (> 10 tests/100 beds/day), cascades were most effective, with a 19-36% probability of detecting outbreaks prior to any nosocomial transmission, and 26-46% prior to first onset of COVID-19 symptoms. Conversely, at low capacity (< 2 tests/100 beds/day), group testing strategies detected outbreaks earliest. Pooling randomly selected patients in a daily group test was most likely to detect outbreaks prior to first symptom onset (16-27%), while pooling patients and staff expressing any COVID-like symptoms was the most efficient means to improve surveillance given resource limitations, compared to the reference requiring only 6-9 additional tests and 11-28 additional swabs to detect outbreaks 1-6 days earlier, prior to an additional 11-22 infections. CONCLUSIONS: COVID-19 surveillance is challenged by delayed or absent clinical symptoms and imperfect diagnostic sensitivity of standard RT-PCR tests. In our analysis, group testing was the most effective and efficient COVID-19 surveillance strategy for resource-limited LTCFs. Testing cascades were even more effective given ample testing resources. Increasing testing capacity and updating surveillance protocols accordingly could facilitate earlier detection of emerging outbreaks, informing a need for urgent intervention in settings with ongoing nosocomial transmission.

Potential for reducing inappropriate antibiotic prescribing in English primary care.

Objectives: To identify and quantify inappropriate systemic antibiotic prescribing in primary care in England, and ultimately to determine the potential for reduction in prescribing of antibiotics. Methods: Primary care data from 2013-15 recorded in The Health Improvement Network (THIN) database were used. Potentially inappropriate prescribing events in the database were identified by: (i) comparing prescribing events against treatment guidelines; (ii) comparing actual proportions of consultations resulting in prescription for a set of conditions with the ideal proportions derived from expert opinion; and (iii) identifying high prescribers and their number of prescriptions above an age- and body-system-specific benchmark. Results: Applying the most conservative assumptions, 8.8% of all systemic antibiotic prescriptions in English primary care were identified as inappropriate, and in the least conservative scenario 23.1% of prescriptions were inappropriate. All practices had non-zero reduction potentials, ranging from 6.4% to 43.5% in the middle scenario. The four conditions that contributed most to inappropriate prescribing were sore throat (23.0% of identified inappropriate prescriptions), cough (22.2%), sinusitis (7.6%) and acute otitis media (5.7%). One-third of all antibiotic prescriptions lacked an informative diagnostic code. Conclusions: This work demonstrates (i) the existence of substantial inappropriate antibiotic prescribing and (ii) poor diagnostic coding in English primary care. All practices (not just the high prescribers) should engage in efforts to improve antimicrobial stewardship. Better diagnostic coding, more precise prescribing guidelines and a deeper understanding of appropriate long-term uses of antibiotics would allow identification of further potential for reductions.

Explaining variation in antibiotic prescribing between general practices in the UK.

Objectives: Primary care practices in England differ in antibiotic prescribing rates, and, anecdotally, prescribers justify high prescribing rates based on their individual case mix. The aim of this paper was to explore to what extent factors such as patient comorbidities explain this variation in antibiotic prescribing. Methods: Primary care consultation and prescribing data recorded in The Health Improvement Network (THIN) database in 2013 were used. Boosted regression trees (BRTs) and negative binomial regression (NBR) models were used to evaluate associations between predictors and antibiotic prescribing rates. The following variables were considered as potential predictors: various infection-related consultation rates, proportions of patients with comorbidities, proportion of patients with inhaled/systemic corticosteroids or immunosuppressive drugs, and demographic traits. Results: The median antibiotic prescribing rate was 65.6 (IQR 57.4-74.0) per 100 registered patients among 348 English practices. In the BRT model, consultation rates had the largest total relative influence on antibiotic prescribing rate (53.5%), followed by steroid and immunosuppressive drugs (31.6%) and comorbidities (12.2%). Only 21% of the deviance could be explained by an NBR model considering only comorbidities and age and gender, whereas 57% of the deviance could be explained by the model considering all variables. Conclusions: The majority of practice-level variation in antibiotic prescribing cannot be explained by variation in prevalence of comorbidities. Factors such as high consultation rates for respiratory tract infections and high prescribing rates for corticosteroids could explain much of the variation, and as such may be considered in determining a practice's potential to reduce prescribing.

Actual versus 'ideal' antibiotic prescribing for common conditions in English primary care.

Objectives: Previous work based on guidelines and expert opinion identified 'ideal' prescribing proportions-the overall proportion of consultations that should result in an antibiotic prescription-for common infectious conditions. Here, actual condition-specific prescribing proportions in primary care in England were compared with ideal prescribing proportions identified by experts. Methods: All recorded consultations for common infectious conditions (cough, bronchitis, exacerbations of asthma or chronic obstructive pulmonary disease, sore throat, rhinosinusitis, otitis media, lower respiratory tract infection, upper respiratory tract infection, influenza-like illness, urinary tract infection, impetigo, acne, gastroenteritis) for 2013-15 were extracted from The Health Improvement Network (THIN) database. The proportions of consultations resulting in an antibiotic prescription were established, concentrating on acute presentations in patients without relevant comorbidities. These actual prescribing proportions were then compared with previously established 'ideal' proportions by condition. Results: For most conditions, substantially higher proportions of consultations resulted in an antibiotic prescription than was deemed appropriate according to expert opinion. An antibiotic was prescribed in 41% of all acute cough consultations when experts advocated 10%. For other conditions the proportions were: bronchitis (actual 82% versus ideal 13%); sore throat (actual 59% versus ideal 13%); rhinosinusitis (actual 88% versus ideal 11%); and acute otitis media in 2- to 18-year-olds (actual 92% versus ideal 17%). Substantial variation between practices was found. Conclusions: This work has identified substantial overprescribing of antibiotics in English primary care, and highlights conditions where this is most pronounced, particularly in respiratory tract conditions.

Defining the appropriateness and inappropriateness of antibiotic prescribing in primary care.

Objectives: To assess the appropriateness of prescribing systemic antibiotics for different clinical conditions in primary care, and to quantify 'ideal' antibiotic prescribing proportions in conditions for which antibiotic treatment is sometimes but not always indicated. Methods: Prescribing guidelines were consulted to define the appropriateness of antibiotic therapy for the conditions that resulted in antibiotic prescriptions between 2013 and 2015 in The Health Improvement Network (THIN) primary care database. The opinions of subject experts were then formally elicited to quantify ideal antibiotic prescribing proportions for 10 common conditions. Results: Of the antibiotic prescriptions in THIN, 52.5% were for conditions that could be assessed using prescribing guidelines. Among these, the vast majority of prescriptions (91.4%) were for conditions where antibiotic appropriateness is conditional on patient-specific indicators. Experts estimated low ideal prescribing proportions in acute, non-comorbid presentations of many of these conditions, such as cough (10% of patients), rhinosinusitis (11%), bronchitis (13%) and sore throat (13%). Conversely, antibiotics were believed to be appropriate in 75% of non-pregnant women with non-recurrent urinary tract infection. In impetigo and acute exacerbation of chronic obstructive pulmonary disease, experts clustered into distinct groups that believed in either high or low prescribing. Conclusions: In English primary care, most antibiotics are prescribed for conditions that only sometimes require antibiotic treatment, depending on patient-specific indicators. Experts estimated low ideal prescribing proportions in many of these conditions. Incomplete prescribing guidelines and disagreement about prescribing in some conditions highlight further research needs.

Antibiotics in primary care in England: which antibiotics are prescribed and for which conditions?

Objectives: To analyse antibiotic prescribing behaviour in English primary care with particular regard to which antibiotics are prescribed and for which conditions. Methods: Primary care data from 2013-15 recorded in The Health Improvement Network (THIN) database were analysed. Records with a prescription for systemic antibiotics were extracted and linked to co-occurring diagnostic codes, which were used to attribute prescriptions to clinical conditions. We further assessed which antibiotic classes were prescribed and which conditions resulted in the greatest share of prescribing. Results: The prescribing rate varied considerably among participating practices, with a median of 626 prescriptions/1000 patients (IQR 543-699). In total, 69% of antibiotic prescriptions (n = 3 156 507) could be linked to a body system and/or clinical condition. Of these prescriptions, 46% were linked to conditions of the respiratory tract, including ear, nose and throat (RT/ENT); leading conditions within this group were cough symptoms (22.7%), lower respiratory tract infection (RTI) (17.9%), sore throat (16.7%) and upper RTI (14.5%). After RT/ENT infections, infections of the urogenital tract (22.7% of prescriptions linked to a condition) and skin/wounds (16.4%) accounted for the greatest share of prescribing. Penicillins accounted for 50% of all prescriptions, followed by macrolides (13%), tetracyclines (12%) and trimethoprim (11%). Conclusions: The majority of antibiotic prescriptions in English primary care were for infections of the respiratory and urinary tracts. However, in almost one-third of all prescriptions no clinical justification was documented. Antibiotic prescribing rates varied substantially between practices, suggesting that there is potential to reduce prescribing in at least some practices.

Understanding the gender gap in antibiotic prescribing: a cross-sectional analysis of English primary care.

OBJECTIVES: To explore the causes of the gender gap in antibiotic prescribing, and to determine whether women are more likely than men to receive an antibiotic prescription per consultation. DESIGN: Cross-sectional analysis of routinely collected electronic medical records from The Health Improvement Network (THIN). SETTING: English primary care. PARTICIPANTS: Patients who consulted general practices registered with THIN between 2013 and 2015. PRIMARY AND SECONDARY OUTCOME MEASURES: Total antibiotic prescribing was measured in children (<19 years), adults (19-64 years) and the elderly (65+ years). For 12 common conditions, the number of adult consultations was measured, and the relative risk (RR) of being prescribed antibiotics when consulting as female or with comorbidity was estimated. RESULTS: Among 4.57 million antibiotic prescriptions observed in the data, female patients received 67% more prescriptions than male patients, and 43% more when excluding antibiotics used to treat urinary tract infection (UTI). These gaps were more pronounced in adult women (99% more prescriptions than men; 69% more when excluding UTI) than in children (9%; 0%) or the elderly (67%; 38%). Among adults, women accounted for 64% of consultations (62% among patients with comorbidity), but were not substantially more likely than men to receive an antibiotic prescription when consulting with common conditions such as cough (RR 1.01; 95% CI 1.00 to 1.02), sore throat (RR 1.01, 95% CI 1.00 to 1.01) and lower respiratory tract infection (RR 1.00, 95% CI 1.00 to 1.01). Exceptions were skin conditions: women were less likely to be prescribed antibiotics when consulting with acne (RR 0.67, 95% CI 0.66 to 0.69) or impetigo (RR 0.85, 95% CI 0.81 to 0.88). CONCLUSIONS: The gender gap in antibiotic prescribing can largely be explained by consultation behaviour. Although in most cases adult men and women are equally likely to be prescribed an antibiotic when consulting primary care, it is unclear whether or not they are equally indicated for antibiotic therapy.

Modelling the evolution of HIV-1 virulence in response to imperfect therapy and prophylaxis.

Average HIV-1 virulence appears to have evolved in different directions in different host populations since antiretroviral therapy first became available, and models predict that HIV drugs can select for either higher or lower virulence, depending on how treatment is administered. However, HIV virulence evolution in response to "leaky" therapy (treatment that imperfectly suppresses viral replication) and the use of preventive drugs (pre-exposure prophylaxis) has not been explored. Using adaptive dynamics, we show that higher virulence can evolve when antiretroviral therapy is imperfectly effective and that this evolution erodes some of the long-term clinical and epidemiological benefits of HIV treatment. The introduction of pre-exposure prophylaxis greatly reduces infection prevalence, but can further amplify virulence evolution when it, too, is leaky. Increasing the uptake rate of these imperfect interventions increases selection for higher virulence and can lead to counterintuitive increases in infection prevalence in some scenarios. Although populations almost always fare better with access to interventions than without, untreated individuals could experience particularly poor clinical outcomes when virulence evolves. These findings predict that antiretroviral drugs may have underappreciated evolutionary consequences, but that maximizing drug efficacy can prevent this evolutionary response. We suggest that HIV virulence evolution should be closely monitored as access to interventions continues to improve.