ProACT was a patient-level, 1:1 randomized trial in 14 hospitals in the United States. We enrolled 1656 adult patients presenting with an initial diagnosis of acute lower respiratory tract infection. Treating clinicians estimated the likelihood of bacterial etiology before any experimental actions. Patients randomly received usual care (no other interventions or guides) or an intervention where the treating clinicians received procalcitonin results and an antibiotic use guideline with graded recommendations based on four tiers of procalcitonin levels. The guideline used the same cutoff values as in previous trials and as approved by the U.S. Food and Drug Administration (antibiotics strongly discouraged for procalcitonin levels < 0.1 mcg/L, discouraged for levels 0.1 to 0.25 mcg/L, recommended for levels > 0.25 to 0.5 mcg/L, and strongly recommended for levels > 0.5 mcg/L). On discharge, we provided patients with a letter for their primary care provider that included their last procalcitonin assay result, a trial synopsis, and the procalcitonin guideline. The primary outcome was total antibiotic exposure to day 30 and data were obtained through chart review and by telephone calls at days 15 and 30 made by coordinating center staff who were unaware of the treatment-group assignments. Using baseline characteristics and published criteria, we categorized the initial diagnosis of lower respiratory tract infection into final diagnoses of acute exacerbation of chronic obstructive pulmonary disease (COPD), asthma exacerbation, acute bronchitis, community-acquired pneumonia, and other.
Statistical analyses
To evaluate the impact of adherence on the primary outcome of antibiotic-days by day 30, we varied adherence between 65% and 100% and determined which threshold adherence rate for low procalcitonin levels (≤ 0.25 mcg/L) in the intervention arm could have allowed rejection of the null hypothesis (no difference between groups) at a significance level of 0.05. The usual care control arm (n = 830) was used for comparison and was not altered in simulations. In the procalcitonin intervention arm we identified all patients who had guideline nonadherent antibiotic prescriptions. To examine 100% adherence, we set antibiotic use to zero at each timepoint when procalcitonin levels were under guideline cutoffs in all intervention arm patients (n = 826). To examine adherence rates < 100%, we simulated an intervention arm population of the same size (n = 826) with the given adherence rate. To account for variability in sampling, we present the mean results of repeating this sampling process 1000 times.
To create each simulated intervention arm population we first included all patients who had only guideline-adherent antibiotic prescriptions. We then used random selection without replacement in R version 3.6 to sample the requisite number of patients with nonadherent prescriptions needed to simulate the given adherence rate. For each of these sampled patients we set their antibiotic use to zero for the timepoints where their procalcitonin levels were below the guideline cutoff. For the patients with nonadherent prescriptions who were not selected and for the patients with only guideline-adherent prescriptions we used their observed antibiotic-days. As with the original report of the ProACT trial, we compared the mean number of antibiotic-days between usual care and intervention groups using two-sample t-tests, following an intention-to-treat approach with multiple imputation to account for patients lost to follow-up after discharge.
Our primary analysis focused on the protocol period in the hospital including both emergency department (ED) and inpatient, where clinicians received procalcitonin results. In sensitivity analyses, we also determined the threshold adherence rate for rejection of the null hypothesis for adherence only in the emergency department, or in the hospital (ED and inpatient) for only for the lowest procalcitonin levels (< 0.1 mcg/L). These analyses addressed whether hospital clinician adherence to procalcitonin guidance only in the emergency department, or only for the lowest procalcitonin levels, could be sufficient to reduce antibiotic use.
We also determined the threshold rate for continued guideline adherence after discharge. For this analysis, we selected patients whose last procalcitonin level prior to discharge was low and who did not receive antibiotics on the day this procalcitonin level was reported. This analysis addressed whether avoidance of overruling in the outpatient setting of initial decisions to withhold antibiotics could be sufficient to reduce antibiotic use.
To explore the second question of how challenging it would be to achieve adherence, we examined the characteristics of patients presenting with a low procalcitonin who were initially prescribed antibiotics on presentation in the emergency department. We then grouped these patients into low, medium, and high risk of illness severity or bacterial infection based on published risk criteria and the treating clinician’s estimate of the likelihood of bacterial etiology [7,8,9]. Our intent was to characterize patient scenarios where the decision to withhold antibiotics would be difficult (high risk), reasonable (low risk), or intermediate (medium risk).
We categorized as high risk all patients with a new infiltrate on chest imaging or > 1 risk criterion of illness severity or bacterial infection. We defined risk criteria based on the systemic inflammatory response syndrome (SIRS), quick Sepsis Related Organ Failure Assessment (qSOFA), the Center for Medicare and Medicaid Services SEP-1 criteria, baseline demographics (age, Charlson comorbidity score), clinician estimate of likelihood of bacterial etiology, oxygen saturation, and hospitalization status [7,8,9]. We limited low risk patients to only those without a new chest infiltrate and with no risk criteria. Thus, we deemed only younger patients with no chest imaging infiltrate, discharged home from the emergency department, with normal vital signs, mental status and oxygen saturation, lower comorbidity burden, and lower clinician estimate of likelihood of bacterial etiology as low risk. We considered medium risk all other patients—i.e., patients without a new chest infiltrate and with at most 1 risk criterion (Appendix Table 3).