Effect of Antimicrobial Therapy on Respiratory Hospitalization or Death in Adults With Idiopathic Pulmonary FibrosisThe CleanUP-IPF Randomized Clinical Trial

; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; for the CleanUP-IPF Investigators of the Pulmonary Trials Cooperative

JAMA. 2021;325(18):1841-1851. doi:10.1001/jama.2021.4956

Key Points

Question  Does antimicrobial therapy in addition to usual care improve clinical outcomes in patients with idiopathic pulmonary fibrosis?

Findings  In this pragmatic randomized clinical trial that included 513 adults with idiopathic pulmonary fibrosis the addition of co-trimoxazole (trimethoprim-sulfamethoxazole) or doxycycline to usual care compared with usual care alone resulted in a rate of first nonelective respiratory hospitalization or death of 20.4 vs 18.4 events per 100 person-years, a difference that was not statistically significant.

Meaning  Among adults with idiopathic pulmonary fibrosis, addition of co-trimoxazole or doxycycline compared with usual care did not significantly improve the time to nonrespiratory hospitalization or death.

Abstract

Importance  Alteration in lung microbes is associated with disease progression in idiopathic pulmonary fibrosis.

Objective  To assess the effect of antimicrobial therapy on clinical outcomes.

Design, Setting, and Participants  Pragmatic, randomized, unblinded clinical trial conducted across 35 US sites. A total of 513 patients older than 40 years were randomized from August 2017 to June 2019 (final follow-up was January 2020).

Interventions  Patients were randomized in a 1:1 allocation ratio to receive antimicrobials (n = 254) or usual care alone (n = 259). Antimicrobials included co-trimoxazole (trimethoprim 160 mg/sulfamethoxazole 800 mg twice daily plus folic acid 5 mg daily, n = 128) or doxycycline (100 mg once daily if body weight <50 kg or 100 mg twice daily if ≥50 kg, n = 126). No placebo was administered in the usual care alone group.

Main Outcomes and Measures  The primary end point was time to first nonelective respiratory hospitalization or all-cause mortality.

Results  Among the 513 patients who were randomized (mean age, 71 years; 23.6% women), all (100%) were included in the analysis. The study was terminated for futility on December 18, 2019. After a mean follow-up time of 13.1 months (median, 12.7 months), a total of 108 primary end point events occurred: 52 events (20.4 events per 100 patient-years [95% CI, 14.8-25.9]) in the usual care plus antimicrobial therapy group and 56 events (18.4 events per 100 patient-years [95% CI, 13.2-23.6]) in the usual care group, with no significant difference between groups (adjusted HR, 1.04 [95% CI, 0.71-1.53; P = .83]. There was no statistically significant interaction between the effect of the prespecified antimicrobial agent (co-trimoxazole vs doxycycline) on the primary end point (adjusted HR, 1.15 [95% CI 0.68-1.95] in the co-trimoxazole group vs 0.82 [95% CI, 0.46-1.47] in the doxycycline group; P = .66). Serious adverse events occurring at 5% or greater among those treated with usual care plus antimicrobials vs usual care alone included respiratory events (16.5% vs 10.0%) and infections (2.8% vs 6.6%); adverse events of special interest included diarrhea (10.2% vs 3.1%) and rash (6.7% vs 0%).

Conclusions and Relevance  Among adults with idiopathic pulmonary fibrosis, the addition of co-trimoxazole or doxycycline to usual care, compared with usual care alone, did not significantly improve time to nonelective respiratory hospitalization or death. These findings do not support treatment with these antibiotics for the underlying disease.

Trial Registration  ClinicalTrials.gov Identifier: 

Introduction

, and is associated with disease progression and a local and systemic immune response,, potentially contributing to acute exacerbations, hospitalizations, and decreased survival. Although initial randomized trial data suggested improved outcomes with co-trimoxazole (trimethoprim-sulfamethoxazole) therapy in patients with fibrosing interstitial lung diseases, a large placebo-controlled trial failed to document an improvement in clinical outcomes with this agent. Preliminary data suggested that doxycycline can improve outcomes, and inhibit metalloproteinases in patients with IPF. Because antimicrobial therapy has been suggested to favorably alter the lung microbial community in other chronic disorders,, this study was designed to address the hypothesis that an antimicrobial therapeutic strategy, with either co-trimoxazole or doxycycline, reduces the risk of nonelective respiratory hospitalization or death among patients with IPF.

Methods

This randomized clinical trial used an approach of 2 potentially effective antimicrobial therapies to simulate clinical practice. This trial was designed with knowledge of the EME-TIPAC (Efficacy and Mechanism Evaluation of Treating Idiopathic Pulmonary Fibrosis With the Addition of Co-trimoxazole) study with the expectation that this therapeutic strategy would better simulate clinical practice, facilitate enrollment, and provide robust results in a timely fashion. The study protocol is provided in ; the statistical analysis plan in ; and the investigator’s assessment of the PRECIS-2 criteria, in eFigure 1 in . The study was approved by institutional review boards at each institution, and each patient provided written informed consent.

Participants

The detailed inclusion and exclusion criteria are enumerated in eTable 1 in . Exclusions were limited and designed to minimize the risk of antimicrobial therapy given the limited in-person follow-up visits. Key exclusions included allergies to both antimicrobial agents, allergy or intolerance to tetracycline and features that increase the risk of co-trimoxazole use, pregnancy, use of an investigational therapy, or concomitant use of immunosuppressive therapy. The National Heart, Lung, and Blood Institute required collection of race, ethnicity, and sex data from participating patients with data self-reported based on fixed categories (American Indian, Asian, Native Hawaiian or other Pacific Islander, Black or African American, White, or other) ().

Randomization

Patients were randomized to the 2 strategies in a 1:1 allocation ratio using a computer-generated algorithm. Prior to randomization, eligible patients and their physician were offered a preference between the co-trimoxazole or doxycycline stratum; the protocol advised co-trimoxazole if possible (). Once patients were randomized their follow-up schedule varied based on their assigned treatment group ().

Intervention

 After randomization, patients assigned to the antimicrobial groups were provided prescription drug vouchers to defray the study drug cost; no placebo agent was provided. Patients were expected to continue on the assigned therapy until the end of the study.

End Points

The primary end point of this study was time to first nonelective, respiratory hospitalization or all-cause mortality. Although forced vital capacity (FVC) has been used as a primary end point in several pivotal trials, the use of clinically relevant parameters has been suggested as a potential advance that is particularly relevant in the setting that clinical trials have been conducted in the postantifibrotic treatment era. Ascertainment of these events included the impression of the investigative clinician (site investigator) and central adjudication group (clinical end point committee) to confirm the clinical cause of a hospitalization or a mortality event. Although the site investigators and participants were not blinded to treatment allocation, the central adjudication group was blinded both to treatment allocation and to the site investigator’s classification of the potential event. The initial central adjudication was provided by a single blinded central committee member. If this assessment agreed with the site investigator’s assessment with respect to elective vs nonelective and respiratory vs nonrespiratory primary cause, the final event classification was achieved. If not, the event was reviewed by a phase 2 adjudication committee that was also blinded to treatment allocation, and a final consensus classification was achieved. The same process was used to confirm the occurrence of deaths and their dates, based on review of submitted medical records.

There were 13 prespecified secondary end points (eTable 2 in ). Due to the COVID-19 pandemic, not all events were able to be adjudicated because some site personnel were pulled for clinical duty, limiting their ability to obtain final clinical data. Due to the timing of the study, none of the patients were affected by clinically identified SARS-CoV-2 infection. The best available classification, used for primary analyses, was the decision of the clinical end point committee, if available, or the clinical site interpretation, if not. Targeted adverse events of special interest such as arrhythmia, diarrhea, hyperkalemia, rash, and vomiting were assessed. To provide simplicity, patients had minimal in-person study-specific visits with data collection coinciding with clinical visits.

Sample Size

The planned sample size of 500 patients was expected to provide adequate power to test the primary hypothesis under a range of event rates. Based on IPFnet data, the primary end point event rates were anticipated to vary depending on the gender, age, and lung physiology (GAP) index scores of the enrolled patients. The statistical power for designs enrolling 500 patients with usual care group events rates varying from 24% to 36% and 12-month treatment effects varying from 30% to 35% are shown in eTable 3 in . The proposed design was expected to provide adequate power under most scenarios. One planned interim analysis for efficacy was scheduled to occur after 300 patients had 12 months of follow-up.

Statistical Analysis

Baseline characteristics were summarized using mean (SDs) or median (interquartile range [IQR]) for continuous variables and frequency in each category for nominal variables. The primary analysis of the primary end point was based on the as-randomized populations. Patients receiving lung transplants during the course of follow-up were censored for all end points at the time of the transplant. The primary comparison between the 2 randomized groups was based on a time-to-event analysis using a multivariable adjusted Cox proportional hazards regression model. The Cox model included treatment group, age, sex, baseline diffusing capacity of lung for carbon monoxide (Dlco), baseline FVC, use of N-acetylcysteine at enrollment, use of nintedanib or pirfenidone at enrollment, and preference of antimicrobial agent prior to randomization. Multiple imputation was performed on age, sex, baseline Dlco, baseline FVC, N-acetylcysteine use at enrollment, use of standard of care medications at enrollment, and preference of antimicrobial agent prior to randomization, using PROC MI, assuming that the data are missing at random with an arbitrary missing pattern; that is, the full conditional specification method was used to generate the 20 imputed data sets. The final inferential results, including the interaction P values, were generated by averaging, using PROC MIANALYZE, the inferential results across the 20 imputed data sets. Hazard ratios (HRs) and 95% CIs were used to summarize the differences between treatment groups. Kaplan-Meier estimates were used to display event rates by treatment group. Participants who were event-free (ie, patients without any respiratory hospitalization or death event at the time of analysis) were censored at their last visit or lung transplant. To assess the interaction of randomized treatment and prespecified antimicrobial agent a variable was added to the model. Departures from the Cox model assumptions were assessed by the examination of the martingale residuals. A post hoc analysis was applied using robust standard errors to account for a possible clustering effect at the enrolling site level. Because of the potential for type I error due to multiple comparisons, findings for analyses of secondary end points should be interpreted as exploratory. The significance threshold was 2-sided at .05. All analyses were performed using SAS version 9.4 software (SAS Institute Inc).

Results

The data and safety monitoring board (DSMB) met on December 18, 2019, to review the first planned efficacy analysis for CleanUP-IPF and unanimously voted to terminate the study early for futility. The determination was based on the low likelihood of the study demonstrating a statistically significant positive effect of the intervention. Neither the protocol nor the DSMB statistical analysis plan contained a formal strategy for a futility analysis. Although no clear harm signal was seen, the DSMB noted a higher rate of the primary outcome, mortality, and serious adverse events in the intervention group that did not reach statistical significance (eAppendix ).

Participants

A total of 513 patients (mean age, 71 years; 21.6% women) at 35 US clinical centers were randomized from August 2017 to June 2019 to receive usual care plus antimicrobials (128, co-trimoxazole; 126 doxycycline) or to usual care alone (259) and were followed up through January 2020 (). Of the participants, 336 (65.5%) were older men with moderate to severe IPF with a high frequency of comorbid conditions, particularly cardiovascular diseases, diabetes, and gastroesophageal reflux disease ( and ). The 2 treatment groups were well matched. Of the 512 study patients with available data, 466 (91%) were taking concurrent antifibrotic medications. Missing information for key baseline variables was no more than 6% and was more than 2% in only 2 baseline variables. Twelve patients were lost to follow-up.

Study Conduct

The study recruited ahead-of-scheduled projections (eFigure 2 in ), with a median follow-up of 12.7 months. Fifty-three patients (10%) withdrew consent (); 49.5% of study patients were followed up for more than 12 months. One patient crossed over from usual care to usual care plus antimicrobial therapy. Within the antimicrobial therapy group, 18 patients switched from co-trimoxazole to doxycycline and 1 from doxycycline to co-trimoxazole therapy during the course of the study.

Adherence to study medication during the course of the study for the 7 days prior to assessment was 78.3% among the 201 patients who reported ) at 1 month, 66.2% at 6 months among 163 patients who reported, and 47.2% at 12 months among 163 patients who reported). There was a statistically significant difference in the withdrawal rates, with more patients withdrawing from the usual care plus antimicrobial therapy: 49 of 254 (19.3%) vs 39 of 259 (15.1%; P = .05) (eFigure 3 in ).

Primary End Point

The time to event end point data, as defined by the clinical end point committee or clinical site, is presented in eFigure 4 and eTable 4 in , and its components, are enumerated in eTable 5 in . The total number of events were quantitatively similar by the committee, the clinical site, or the best available data. For adjudication of respiratory related nonelective hospitalizations the site determination result vs the clinical end point committee result were found to strongly agree with a κ of 0.78 (95% CI, 0.68-0.88) and a low and balanced disagreement rates of 10.3% when the site said yes and the clinical end point committee said no and 11.4% when the clinical end point committee said no and the site said yes.

When the database locked, a total of 108 primary end point events occurred: 52 events (20.4 events per 100 patient-years [95% CI, 14.8-25.9]) in the usual care plus antimicrobial therapy group and 56 events (18.4 events per 100 patient-years [95% CI, 13.2-23.6]) in the usual care group. The mean follow-up time was 13.1 months. There was no significant effect of antimicrobial therapy after multivariable adjustment (HR, 1.04 [95% CI, 0.71-1.53]; , ; and eTable 6 in ) for time to the composite primary end point. A post hoc analysis accounting for the clustering of patients within the enrolling clinical site yielded a similar result (eTable 6 in ). An examination of the martingale residuals did not reveal any departures from the assumed model. There was no significant interaction between the effect of the prespecified antimicrobial agent on the primary end point (adjusted HR for co-trimoxazole, 1.15 [95% CI, 0.68-1.95] vs doxycycline, 0.82 [95% CI, 0.46-1.47]; P = .66) (eFigure 5 in ). The overall number of events rose by each GAP stage (eTable 7 in ).

Secondary End Points

In covariate adjusted analyses, there was no statistically significant difference in the time to death between the antimicrobial treatment and usual care (13.5 vs 11.5 events per 100 patient-years; HR, 1.11 [95% CI; 0.70-1.78]) or to the time to the first nonelective respiratory hospitalization (14.1 vs 10.2 hospitalizations per 100 patient-years; HR, 1.34 [95% CI, 0.82-2.17]) (, , and eTable 6 in ). Similarly, neither antimicrobial treatment added to usual care had a significant effect on the time to death (interaction P value = .28). Co-trimoxazole had an HR of 1.48 (95% CI, 0.78-2.81) and doxycycline, HR of 0.74 (95% CI, 0.37-1.49) on the time to the first nonelective respiratory hospitalization with an HR of 1.40 (95% CI, 0.72-2.75) for co-trimoxazole and an HR of 1.10 (95% CI, 0.53-2.26) for doxycycline (interaction P value = .79).

The time to the first nonelective all-cause hospitalization did not differ in statistical significance between antimicrobial therapy (22.6 hospitalizations per 100-patient years) and usual care (15.8 hospitalizations per 100 patient-years; HR, 1.36 [95% CI, 0.91- 2.01]) nor in the total number of nonelective all-cause hospitalizations ().

The total number of nonelective all-cause hospitalizations showed slightly higher rates in the usual care plus antimicrobial groups than in usual care alone group (26.4 vs 17.5 hospitalizations per 100 patient-years; rate ratio; 1.44 [95% CI, 1.00-2.07]). There were no statistically significant differences between usual care plus antimicrobial therapy vs usual care alone in the change between randomization and 12 months in FVC, Dlco, or patient-reported outcome parameters (, eTable 8, and eFigure 6 in ).

Adverse Events

Numerically more respiratory serious adverse events in patients treated with usual care plus antimicrobial therapy were observed, whereas fewer infections were observed (). There were expected differences in prespecified adverse events of special interest. These included more episodes of diarrhea, vomiting, rash, arrhythmias, and hyperkalemia in the antimicrobial therapy group. Similar patterns of serious adverse events and prespecified events of special interest were observed for the co-trimoxazole and doxycycline cohorts (eTable 9 in ).

Discussion

Among adults with IPF, adding co-trimoxazole or doxycycline to usual care did not significantly improve time to nonelective respiratory hospitalization or death compared with usual care alone. These findings do not support treatment with these antibiotics for the underlying disease.

Neither EME-TIPAC nor this study supported a significant effect of either co-trimoxazole on the primary end point or on numerous secondary end points. Although both studies used different designs, they had a very similar composite of clinical outcomes that was thought to be clinically relevant and more reflective of the anticipated mechanism of action of antimicrobial therapy in contrast to the more traditional longitudinal change in FVC. As expected this study’s patient population was rapidly enrolled, was physiologically less severe, experienced greater comorbidity, and exhibited lower adherence to the study therapy. Despite these differences, the withdrawal rate was identical and the primary end point for co-trimoxazole therapy was similar between the 2 studies.

Increased gastrointestinal adverse events, particularly those treated with doxycycline, were identified. Rash and hyperkalemia were more frequently seen with co-trimoxazole therapy. These adverse events are reflective of those reported in other settings, were not serious in nature, and were similar to reactions to co-trimoxazole, noted in the EME-TIPAC study. Nevertheless, there was a numerical imbalance in respiratory serious adverse events but not infections with antimicrobial therapy. The latter observation suggests a possible disconnect between IPF progression and pneumonia, with the former potentially being less likely to be associated with bacterial infection.

This study extends the EME-TIPAC findings by suggesting that a broader antimicrobial therapy strategy was not effective in improving clinical outcomes. Several investigations have demonstrated an altered lung microbial community in patients with IPF that correlates closely with clinical outcomes, and is associated with augmented local and systemic immune responses., Robust murine data support the association of the lung microbial community on increased fibroproliferative response,; however, antimicrobial therapy or lack of microbial populations in germ-free mice has ameliorated this association., The findings may reflect the complex nature of the lung microbial community in chronic lung disease and the approach to define causality by altering dysbiosis with antimicrobial therapy, including significant interpatient and intrapatient heterogeneity in the microbiome and its associated host response. particularly as there appears to be a core resistome in chronic lung diseases. The results of this study, in conjunction with those of EME-TIPAC investigators, do not support the use of co-trimoxazole to modify disease progression in patients with IPF but do not negate the potential effects of other agents whose preclinical or preliminary clinical work suggest an antifibrotic and/or clinical effect.

Because this study did not measure the number and type of lung microbes nor the direct effect of antibiotics on these microbes, this study could not exclude that antimicrobials may work in select subsets of patients with IPF experiencing increased dysbiosis. Other approaches for manipulation of the microbiome from fecal transplant, phage therapy, or selective immunization, among others, may prove successful. This study collected blood for DNA and RNA, along with buccal and fecal swabs for future analyses, including candidate genetic association studies and exploratory genome and transcriptome-wide studies.

Limitations

This study has several limitations. First, there was lack of placebo control and less intrusive adherence monitoring that may have limited the ability to robustly define treatment effect as was suggested in a preliminary study. Second, there was not a bronchoalveolar lavage or other identification of lung dysbiosis for patient selection. Third, the IPF diagnosis was established by the site investigator rather than by investigators in a centralized location, so it is possible that enrolled patients did not strictly meet current diagnostic criteria. Fourth, it is possible that the differential in smoking rates observed between treatment groups influenced levels of lung dysbiosis potentially influencing the lack of observed treatment effect.

Conclusions

Among adults with idiopathic pulmonary fibrosis, the addition of co-trimoxazole or doxycycline to usual care, compared with usual care alone, did not significantly improve time to nonelective respiratory hospitalization or death. These findings do not support treatment with these antibiotics for the underlying disease.

Article Information

Corresponding Author: Fernando J. Martinez, MD, MS, 1305 York Ave, PO Box 96, Room Y-1059, New York, NY 10021 ([email protected]).

Accepted for Publication: March 16, 2021.

Author Contributions: Drs Martinez and Anstrom had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Drs Anstrom and Noth are co–senior authors.

Concept and design: Martinez, Flaherty, Wisniewski, Sciurba, Raghu, D. Kim, Criner, Scholand, Anstrom, Noth.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Martinez, Yow, H. Kim, Anstrom, Noth.

Critical revision of the manuscript for important intellectual content: Martinez, Flaherty, Snyder, Durheim, Wisniewski, Sciurba, Raghu, Brooks, D. Kim, Dilling, Criner, H. Kim, Belloli, Nambiar, Scholand, Anstrom, Noth.

Statistical analysis: Martinez, Yow, Wisniewski, D. Kim, Anstrom.

Obtained funding: Martinez, Sciurba, Noth.

Administrative, technical, or material support: Martinez, Snyder, Wisniewski, Sciurba, Criner, Nambiar, Scholand, Anstrom, Noth.

Supervision: Martinez, Snyder, Sciurba, Criner, Nambiar, Anstrom, Noth.

Conflict of Interest Disclosures: Dr Martinez reported serving on the COPD advisory boards of AstraZeneca, Bayer, Boehringer Ingelheim, Chiesi, Sunovion, CSL Behring, Gala, GlaxoSmithKline, Novartis, Polarean, ProterrixBio, Sanofi/Regeneron, Teva, and Verona; COPD study steering committees of AstraZeneca, GlaxoSmithKline, Chiesi, and Sanofi; Interstitial Lung Disease advisory boards or consulting for Abbvie, Boehringer Ingelheim, Bristol Myers Squibb, Bridge Biotherapeutics, CSL Behring, DevPro, Genentech, IQVIA, Sanofi, Shionogi, twoXAR, United Therapeutics, Veracyte, and Zambon. Interstitial Lung Disease study steering committees for Afferent/Merck, Bayer, Biogen, Boehringer Ingelheim, Gilead, Patara/Respivant, ProMedior/Roche, and Veracyte; the data and safety monitoring boards for Biogen, Boehringer Ingelheim, Genentech, GlaxoSmithKline, and Medtronic; advisory boards/consultation on COVID-19 related topics for AstraZeneca, Boehringer Ingelheim, GlaxoSmithKline, and Raziel; receiving personal fees for continuing medical education from the Academy for Continuing Healthcare Learning, CME Outfitters, Dartmouth University, Integritas, Integrity Communications, MedScape, Methodist Hospital Brooklyn, Miller Communications, National Association for Continuing Education/Haymarket, PeerView Communications, Physicians Education Resource Program, Projects in Knowledge, UpToDate, Vindico, and WebMD; grants from the National Institutes of Health (NIH); and personal fees for serving as deputy editor of the American Journal of Respiratory and Critical Care Medicine. Dr Durheim reported receiving grants from the National Heart, Lung, and Blood Institute (NHLBI) and personal fees from Boehringer Ingelheim and Roche. Dr Sciurba reported receiving institutional grants from the NIH. Dr Raghu reported receiving institutional grants from the NIH and consulting fees from Boehringer Ingelheim and Roche-Genentech. Dr Brooks reported receiving grants from NIH. Dr Dilling reported receiving grants from University of Pittsburgh Pulmonary Trials Cooperative (PTC), personal fees from the Genentech and Boehringer Ingelheim, grant support from Boehringer Ingelheim, Nitto Denko Corp, Galapagos NV, Gilead Sciences, Bellerophon Pulse Technologies, and Duke Clinical Research Institute. Dr Criner reported receiving grants from Boehringer Ingelheim, personal fees from Boehringer Ingelheim, grants from Galapagos and Patara. Dr Belloli reported receiving grants from the NIH. Dr Nambiar reported receiving grants from the University of Pittsburgh, the NHLBI, Roche/Genentech, Pulmonary Fibrosis Foundation, Nitto Denko, FibroGen, and Galapagos; and personal fees from Boehringer Ingelheim, Roche/Genentech, and Veracyte. Dr Scholand reported receiving personal fees from Genentech, Boerhinger Ingelheim, Veracyte, and United Therapeutics. Dr Anstrom reported receiving grants from Merck, Bayer, and the NIH. Dr Noth reported receiving grants from the NIH, Veracyte, and Three Lakes Foundation; personal fees from Boerhinger Ingelheim, Genentech, and Confo and having a patent for TOLLIP in IPF pending. No other disclosures were reported.

Funding/Support: This work was supported by grants U01HL128964 from the NIH/NHLBI, Three Lakes Foundation, IPF Foundation, and Veracyte Inc.

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.

Group Information: A listing of the CleanUP-IPF Investigators of the Pulmonary Trials Cooperative appears in .

Data Sharing Statement: See .

Additional Contributions: We thank Michelle LoPiccolo, MHA, BS, NewYork-Presbyterian Hospital/Weill Cornell Medicine, for editorial assistance.

References

22.QualityMetric Health Outcomes Scoring Software 5.0 User’s Guide. QualityMetric Inc; 2016.

Nguồn://jamanetwork.com/journals/jama/fullarticle/2779827?guestAccessKey=f389242d-ca9d-48c8-bf8e-1ae14ffe029b&utm_source=silverchair&utm_medium=email&utm_campaign=article_alert-jama&utm_content=etoc&utm_term=051121