Sean Ong

Quality-of-Care Indicator Adherence and Mortality Outcomes in MRSA Bacteremia This post hoc analysis of a randomized clinical trial investigates whether patient participation in clinical trials is associated with practitioner adherence to methicillin-resistant Staphylococcus au...

New post-hoc Camera-2 📷 post-hoc analysis out today regarding adherence to quality metrics in MRSA bacteremia.

Compared to similar non-trial population w/ MRSA bacteremia:
-No difference in 🪦, BUT
-non-trial had less following of quality metrics (see ⬇️)

#IDSky

jamanetwork.com/journals/jam...

This study was partly inspired after seeing a slide on @steventong.bsky.social's presentation about SAB trials at ESCMID 2024, after which we connected with the BACSARM and SAFO investigators to put this together. Conferences are a wonderful source of new project ideas and collaborations.

Glad to have the opportunity to collaborate with Spanish colleagues on this. We conducted a pooled Bayesian post-hoc analysis of the BACSARM and SAFO trials evaluating combination therapy with fosfomycin for SAB. The door may not be closed on adjunctive fosfomycin! Justifies need for more RCTs.

This is a fantastic example of a Study Within A Trial (SWAT), incorporating research on RCT methods within a larger clinical trial. We are currently conducting a similar SWAT, SIMPLY-SNAP, which looks at the impact of simplified layered consent on recruitment to SNAP: doi.org/10.1136/bmjo...

Trials Within Trials—Optimizing the Delivery of RCTs In the latest issue of JAMA, Johansen and colleagues1 report the results of a trial that investigated the impact of digital recruitment letter formats on recruitment to a larger clinical trial, the DA...

Thankful for another opportunity to write this editorial with @steventong.bsky.social in @jama.com: jamanetwork.com/journals/jam...

We discuss this trial, which evaluated the impact of different recruitment letter formats on RCT enrolment: jamanetwork.com/journals/jam...

8/ We tried to write this review in an approachable way for clinicians who may not be entirely comfortable with statistics. Hopefully it is useful to some folks.

In a follow-up paper, we applied the different methods described in illustrative post-hoc analyses of BALANCE and CAMERA2. Stay tuned!

7/ The nomenclature in the literature is inconsistent, and often mixes up outcomes, target parameters, and analytic methods. Confusingly, outcomes can be summarized using multiple different target parameters, which can be analyzed using multiple different methods. More than one way to skin a cat :)

6/ Importantly, we should look at whether odds are proportional (known as the PO assumption), or minimally that the direction of effect is consistent across all levels of the ordinal scale. As long as the effect is consistent (as opposed to discordant), the common OR is clinically interpretable.

5/ Parametric approaches can also be used for HCEs, commonly the proportional odds (PO) model. This estimates the common OR, interpreted as the odds a patient in one group has of being at least one step higher on the scale (i.e., having a better outcome) than a patient in the other group.

4/ This results in three summary numbers: total number of wins, losses, and ties. These numbers can be used to calculate different treatment effect estimates: win ratio, win odds, net treatment benefit, NNT, and probabilistic index. (Don't let the math scare you, it's simpler than it looks!)

3/ HCEs can also be analyzed using the generalized pairwise comparisons (GPC) method, where every patient in group 1 is compared to every patient in group 2 in all possible patient-pair combinations. A win is determined for each pair based on the hierarchical ordering of outcome components.

2/ Studies using DOOR are often analyzed using the Mann-Whitney U test, comparing distribution of ranks across groups. This estimates the probability that a randomly selected patient in group 1 has a better outcome to one in group 2 (AKA probabilistic index).

doi.org/10.1016/j.cm...

1/ HCEs combine features of composite and ordinal outcomes: they combine multiple clinical events (composite), but are also hierarchical as they establish a rank order of clinical importance across the different component events. A common example in ID is the DOOR outcome, a specific type of HCE.

Making sense of hierarchical composite endpoints in randomized clinical trials – a primer for infectious disease clinicians and researchers Hierarchical composite endpoints (HCEs) are increasingly being used in infectious diseases research. In this review aimed at educating the clinical reader,

Our review on hierarchical composite endpoints (HCEs) has just been published in @cidjournal.bsky.social:
doi.org/10.1093/cid/...

We provide a conceptual overview of HCEs, explain the different associated target parameters and analytic methods.

@steventong.bsky.social @gurujosh.bsky.social

Venn diagram with one circle representing ordinal endpoints (eg WHO clinical progression score for COVID, modified Rankin scale, Glasgow outcome score, etc), another representing composite endpoints (eg time to death, treatment failure, or disease recurrence), and area of overlap (hierarchical composite endpoints, eg DOOR and extensions as seen in REMAP-CAP).

Making sense of hierarchical composite endpoints in randomized clinical trials – a primer for infectious disease clinicians and researchers

@seanong.bsky.social @gurujosh.bsky.social @steventong.bsky.social and colleagues

Open Access #IDSky #MedSky

academic.oup.com/cid/advance-...

Target Trial Emulation for Antibiotic Use in Acute COVID-19 This study by Pulia and colleagues1 reports results from a large population-based cohort using target trial emulation to examine the associations of antibiotic treatment with clinical outcomes in pati...

Honoured to be invited to write this commentary with @steventong.bsky.social, on an excellent target trial emulation study which showed that antibiotic use was not associated with improved outcomes in non-severe COVID-19.

jamanetwork.com/journals/jam...

#IDSky

🆕 Evaluating BSI in Ontario we found #AMR was associated with a 10% ⬆️ risk of death equivalent to 1.2 deaths/100,000 people/year

These estimates are lower than those in previous literature, which may be due to robust adjustment for confounding

doi.org/10.1093/cid/...
@cidjournal.bsky.social

Evaluating Harms Associated With Prolonged Antibiotic Duration of Therapy in Community-Dwelling Older Adults: A Cohort Study Using Instrumental Variable Analysis Shorter courses of antibiotic therapy are increasingly recommended to reduce antibiotic exposure. However, quantifying the real-world impact of duration of therapy is hindered by bias common in observational studies.

Evaluating Harms Associated With Prolonged Antibiotic Duration of Therapy in Community-Dwelling Older Adults: A Cohort Study Using Instrumental Variable Analysis

#IDSky

Forest plot of 30 day all cause mortality which favors cefazolin (OR=0.73, 95%CI=0.62-0.85)

Forest plot of discontinuation due to adverse effects which favors cefazolin (OR=0.13, 95%CI=0.06-0.27)

Cefazolin vs Antistaphylococcal Penicillins for Treatment of Methicillin-Susceptible Staphylococcus aureus Bacteremia: A Systematic Review & Meta-Analysis

@connorprosty.bsky.social @seanong.bsky.social &c

Cefaz more effective, safer #idsky
www.clinicalmicrobiologyandinfection.com/article/S119... 🔓

It's a reason some people might throw up to say the SNAP results don't apply to their setting... Because we only used (flu)cloxacillin in SNAP. So we did this SR/MA to try and address that.

But at the same time I recognize that every trial/study setting has a unique sociodemographic context, so there isn't a one-size-fits-all solution! And being overly prescriptive and mandating reporting may also be counterproductive. As with most things, a nuanced approach is best.

I'm not sure that is sufficient. We can't assume that trials are representative of the patient population in that hospital catchment area. There are many documented barriers from screening to trial enrolment. IMO it's better to transparently report granular data as much as is feasibly possible.

Thank you! Indeed, lots to improve on how we collect and report these data.

It's generally poor and inconsistent across most fields:
jamanetwork.com/journals/jam...
www.thelancet.com/journals/ecl...
bmjopen.bmj.com/content/12/8...

But to our knowledge, our study is the first to specifically examine the ID literature.

10/ Lastly – we collected a lot more data as part of this massive review of 1343 RCTs. I really have to thank the combined efforts of a multinational team of students and fellows in helping with data extraction. Keep a look out for more secondary papers from this dataset!

9/ Our study shows that reporting of important sociodemographic variables was low, inconsistent, and has not improved over time. Consensus guidelines should outline the minimum set of variables that should be reported in RCTs, and describe the specific settings that require more detailed reporting.

8/ Beyond the presence/absence of reporting, it is also important to ensure quality and consistency of reporting. E.g., 18% of studies that reported ethnicity in our review only reported binary categories (e.g., white vs non-white), which is inadequate in capturing the diversity of most populations.

7/ In multivariable models, we found that subject area, funding source, health equity relatedness, use of a professional medical writer, and trial setting (high income vs LMIC) were independently associated with probability of reporting of ethnicity, education, or socioeconomic status.

6/ Reporting of these variables also differed widely depending on subject area. E.g., RCTs studying viral infections, HIV/STIs, and COVID-19 had the highest reporting of ethnicity, while those studying tropical medicine, critical care, and antimicrobial stewardship/infection control had the lowest.

5/ Almost all trials reported sex and/or gender. However, less than half (49.3%) reported ethnicity, and only a minority reported education level (9.0%), socioeconomic status (9.0%), and rurality (3.9%). Reporting of all these variables did not improve throughout the 10-year study period.