Predictors of Adverse Outcomes in Uncomplicated Lower Respiratory Tract Infections

Thanks for those comments

We have produced an amended table 6 which includes the absolute numbers of those with and without the outcome at each cut point, we will ask this to be added as a further appendix. We agree whilst these can be imputed from the table that providing the figures does ease interpretation in demonstrating the numbers at each level of score from the cohort and the numbers with and without adverse outcome. The likelihood ratios quoted are in effect the stratum specific values, the quoted values can be replicated using the formula provided by Deeks and Altman (1). So these can be used to understand the implications for clinicians of selecting a particular cut point to attribute potential risk of adverse outcome.

1. Deeks JJ, Altman DG: Diagnostic tests 4: likelihood ratios. BMJ 2004, 329(7458):168-169.

Competing interests: None declared

Introduction: The purpose of the study was to help identify individuals with lower respiratory tract infections that are at risk of adverse outcomes in order to treat them appropriately in the primary care setting. The authors acknowledge that respiratory tract infections are one of the most common acute illnesses that present to the primary care office, and are concerned with reducing the number of adverse outcomes by studying and better identifying those individuals who are at greatest risk. The authors touched on the heavy use of antibiotics and a large study that evaluated the effect of their use on symptoms, stating that most clinicians and patients are more concerned about reducing the chance of adverse outcomes versus symptom reduction as a point of relevance.

Methods: Since adverse outcomes with respiratory infections is uncommon overall, a large cohort was expected. The authors recruited a prospective cohort of 28,883 adults from 522 practices, and compiled data from the initial consultations and medical records. Individuals chosen for inclusion in the study needed to have been aged 16 years or older and met the Cochrane review of antibiotics for bronchitis definition of LRTI. Patients were excluded if they any other cause of cough, were unable to fill out a diary, immunocompromised, or presenting repeatedly with the same episode of illness. Of note, a "Clinical Record Form" was developed by general practitioners for the purpose of collecting data such as demographics, descriptors of current illness, examination findings, subjective rating of overall illness severity, and whether or not antibiotics were used. The second component consisting of a note review including imaging, outcomes, comorbidities, and vaccination status. The primary outcome of an adverse event (i.e. hospitalization after the first day or death) and diagnosis of pneumonia from days 2-30 after initial encounter was determined from the records, either through clinical notes or imaging reports. They excluded patients diagnosed within 7 days with pneumonia based on imaging without additional consultation. Our group discussed whether or not the Cochrane review definition of LRTI, i.e. "acute cough (new or worsening cough for less than or equal to 3 weeks) presenting as the main symptom and judged to be infective in origin by the physician" was specific enough for lower respiratory infection, and does not preclude a diagnosis of upper respiratory tract infection, post-nasal drip, etc., and was rather subjective in its determination of infective origin. Additionally, we were not clear on the reasoning to omit cases admitted on day 1. Finally, a "gut feeling" illness severity score made on behalf of the clinicians initially seemed as though it could be improved with greater standardization or training, but it was pointed out amongst our group that it may be left as is in order to not interfere with study evaluation of the current and natural state of the clinicians' training level and professional expertise. The researchers then utilized a generalized linear model to evaluate variables that were predictors of outcome and then assess with logistic regression. The student group agreed that an initial a priori approach to this large of a cohort with a large number of potential and relevant explanatory variables was appropriate.

Results One-hundred and twenty cases of pneumonia plus 34 non-X-ray confirmed cases of the 28,883 study participants were included in the primary analysis. From this, they determined that the frequency of 1 or more serious adverse event was 1.1% overall, or 325 out of 28,846, with 29 (0.1%) deaths, 120 (0.7%) cases of late-onset pneumonia, and 196 (0.7%) hospital admissions. In table 3, the authors reported adjusted risk ratios for severe adverse outcomes of various patient characteristics, symptoms, and exam findings. Our student group shared that we would have liked to have seen a separate table with the deaths removed from the combined adverse outcomes of both death or hospitalization, as both outcomes are vastly different in terms of severity, implications for future care, and cost. Additionally, a future study could be conducted to evaluate what, if any, characteristics were differentiated the patients that died but were not admitted to the hospital. The authors noted that clinical diagnosis of pneumonia is known to be specific but lacking in sensitivity. Missing the potential for treatment prior to death for any reason, either cost of care, lack of severe symptoms, patient choice, etc. could provide compelling analysis with a variety of applications whether it be filling in diagnostic gaps, advocating for increased access to care, or substantiating end-of-life decision-making.

The authors identified 8 independent predictors of adverse outcomes including: certain clinical features (i.e. no coryza, chest pain, fever, severity score greater than 5/10), inherent risk factors (i.e. age 65 years or older, comorbidity), and physical findings (i.e. oxygen saturation less than 95%, low blood pressure). To make the findings clinically relevant, a simple clinical prediction scoring system was based off of the 8 items, each assigned a value of 1. Individuals scoring greater than or equal to 8 had a positive predictive value of 66.7%. The cutoff values to divide groups into "high-, intermediate, and low-risk groups," corresponded to PPVs of 5.7%, 2%, and 0.5% respectively. Our group noticed that although some other features were statistically significant, notably shortness of breath, chills, headache, purulent sputum, and respiratory rate greater than 24/min, they were excluded from the chosen 8 criteria although we could not identify a reason why they were not included in the final clinical scoring system.

Discussion / Conclusion: Overall, our group agreed that the study was robust and applicable to many populations. The purpose of the study seemed to have been simply identifying those with LRTI who are at greater risk of adverse outcomes, but with the eventual goal of seeking out what can be done in a clinical setting to reduce these adverse outcomes. Nevertheless, it is reassuring that overall most (98.9%) of patients will not suffer from either death or hospitalization. It may be obvious to the well-trained and experienced clinician that any particular case may be of increased risk of an adverse outcome based simply on the number of symptoms, findings, and risk factors present, so the study seemed redundant in this regard. However, identifying specific factors that put a patient more at risk, and attaching a risk percentage is helpful. In this sense, the study provides a valuable tool for making decisions about the level aggressiveness of treatment that should be pursued in the outpatient setting.

Our group discussed the possibility of a subgroup analysis, particularly between score cutoffs (greater than or equal to 5, or 8) and antibiotic utilization. The question was posed as to whether these patients fared better if they received antibiotics if they had a higher score from the 8 chosen categories. A similar subgroup analysis could be conducted on vaccine utilization to address whether or not vaccination reduced the likelihood of adverse outcomes. Both analyses would add to evidence-based clinical preventative care by advising clinicians of the benefit or harm of administering vaccines and antibiotics to any particular patient based on their individual set of clinical, exam characteristics. Finally, our group was particularly interested in seeing a separate analysis for both death and hospitalization, as both are different outcomes with very different implications for future research, as discussed previously.

Competing interests: None declared