Derivation of a triage algorithm for chest radiography of community-acquired pneumonia patients in the emergency department

Pulse Oximetry as an Aid to Rule Out Pneumonia among Patients with a Lower Respiratory Tract Infection in Primary Care

Guidelines recommend chest X-rays (CXRs) to diagnose pneumonia and guide antibiotic treatment. This study aimed to identify clinical predictors of pneumonia that are visible on a chest X-ray (CXR+) which could support ruling out pneumonia and avoiding unnecessary CXRs, including oxygen saturation. A secondary analysis was performed in a clinical trial that included patients with suspected pneumonia in Swiss primary care. CXRs were reviewed by two radiologists. We evaluated the association between clinical signs (heart rate > 100/min, respiratory rate ≥ 24/min, temperature ≥ 37.8 °C, abnormal auscultation, and oxygen saturation < 95%) and CXR+ using multivariate analysis. We also calculated the diagnostic performance of the associated clinical signs combined in a clinical decision rule (CDR), as well as a CDR derived from a large meta-analysis (at least one of the following: heart rate > 100/min, respiratory rate ≥ 24/min, temperature ≥ 37.8 °C, or abnormal auscultation). Out of 469 patients from the initial trial, 107 had a CXR and were included in this study. Of these, 26 (24%) had a CXR+. We found that temperature and oxygen saturation were associated with CXR+. A CDR based on the presence of either temperature ≥ 37.8 °C and/or an oxygen saturation level < 95% had a sensitivity of 69% and a negative likelihood ratio (LR-) of 0.45. The CDR from the meta-analysis had a sensitivity of 92% and an LR- of 0.37. The addition of saturation < 95% to this CDR increased the sensitivity (96%) and decreased the LR- (0.21). In conclusion, this study suggests that pulse oximetry could be added to a simple CDR to decrease the probability of pneumonia to an acceptable level and avoid unnecessary CXRs.

Clinical prediction rule to predict pneumonia in adult presented with acute febrile respiratory illness

OBJECTIVE

To derive a clinical prediction rule to predict pneumonia in patients with acute febrile respiratory illness to emergency departments.

METHOD

This was a prospective multicentre study. 537 adults were recruited. Those requiring resuscitation or were hypoxaemic on presentation were excluded. Pneumonia was defined as new onset infiltrates on chest X-ray (CXR), or re-attendance within 7 days and diagnosed clinically as having pneumonia. A predictive model, the Acute Febrile Respiratory Illness (AFRI) rule was derived by logistic regression analysis based on clinical parameters. The AFRI rule was internally validated with bootstrap resampling and was compared with the Diehr and Heckerling rule.

RESULTS

In the 363 patients who underwent CXR, 100 had CXR confirmed pneumonia. There were 7 weighted factors within the ARFI rule, which on summation, gave the AFRI score: age ≥ 65 (1 point), peak temperature within 24 h ≥ 40 °C (2 points), fever duration ≥3 days (2 points), sore throat (-2 points), abnormal breath sounds (1 point), history of pneumonia (1 point) and SpO2 ≤ 96% (1 point). With the bootstrap resampling, the AFRI rule was found to be more accurate than the Diehr and Heckerling rule (area under ROC curve 0.816, 0.721 and 0.566 respectively, p < 0.001). At a cut-off of AFRI≥0, the rule was found to have 95% sensitivity, with a negative predictive value of 97.2%. Using the AFRI score, we found CXR could be avoided for patients having a score of <0.

CONCLUSION

AFRI score could assist emergency physicians in identifying pneumonia patients among all adult patients presented to ED for acute febrile respiratory illness.

Automating Construction of Machine Learning Models With Clinical Big Data: Proposal Rationale and Methods

Background

To improve health outcomes and cut health care costs, we often need to conduct prediction/classification using large clinical datasets (aka, clinical big data), for example, to identify high-risk patients for preventive interventions. Machine learning has been proposed as a key technology for doing this. Machine learning has won most data science competitions and could support many clinical activities, yet only 15% of hospitals use it for even limited purposes. Despite familiarity with data, health care researchers often lack machine learning expertise to directly use clinical big data, creating a hurdle in realizing value from their data. Health care researchers can work with data scientists with deep machine learning knowledge, but it takes time and effort for both parties to communicate effectively. Facing a shortage in the United States of data scientists and hiring competition from companies with deep pockets, health care systems have difficulty recruiting data scientists. Building and generalizing a machine learning model often requires hundreds to thousands of manual iterations by data scientists to select the following: (1) hyper-parameter values and complex algorithms that greatly affect model accuracy and (2) operators and periods for temporally aggregating clinical attributes (eg, whether a patient’s weight kept rising in the past year). This process becomes infeasible with limited budgets.

Objective

This study’s goal is to enable health care researchers to directly use clinical big data, make machine learning feasible with limited budgets and data scientist resources, and realize value from data.

Methods

This study will allow us to achieve the following: (1) finish developing the new software, Automated Machine Learning (Auto-ML), to automate model selection for machine learning with clinical big data and validate Auto-ML on seven benchmark modeling problems of clinical importance; (2) apply Auto-ML and novel methodology to two new modeling problems crucial for care management allocation and pilot one model with care managers; and (3) perform simulations to estimate the impact of adopting Auto-ML on US patient outcomes.

Results

We are currently writing Auto-ML’s design document. We intend to finish our study by around the year 2022.

Conclusions

Auto-ML will generalize to various clinical prediction/classification problems. With minimal help from data scientists, health care researchers can use Auto-ML to quickly build high-quality models. This will boost wider use of machine learning in health care and improve patient outcomes.

Traditional nurse triage vs physician telepresence in a pediatric ED

OBJECTIVES

The objective of the study is to compare traditional nurse triage (TNT) in a pediatric emergency department (PED) with physician telepresence (PTP).

METHODS

This is a prospective 2 × 2 crossover study with random assignment using a sample of walk-in patients seeking care in a PED at a large, tertiary care children's hospital, from May 2012 to January 2013. Outcomes of triage times, documentation errors, triage scores, and survey responses were compared between TNT and PTP. Comparison between PTP to actual treating PED physicians regarding the accuracy of ordering blood and urine tests, throat cultures, and radiologic imaging was also studied.

RESULTS

Paired samples t tests showed a statistically significant difference in triage time between TNT and PTP (P = .03) but no significant difference in documentation errors (P = .10). Triage scores of TNT were 71% accurate, compared with PTP, which were 95% accurate. Both parents and children had favorable scores regarding PTP, and most indicated that they would prefer PTP again at their next PED visit. Physician telepresence diagnostic ordering was comparable with the actual PED physician ordering, showing no statistical differences.

CONCLUSIONS

Using PTP technology to remotely perform triage is a feasible alternative to traditional nurse triage, with no clinically significant differences in time, triage scores, errors, and patient and parent satisfaction.

Time to first antibiotics for pneumonia is not associated with in-hospital mortality

BACKGROUND

Time to first antibiotic (TTFA) is postulated to impact pneumonia mortality. The Joint Commission/Centers for Medicare and Medicaid Services national quality standards previously indicated that TTFA should be <6 h (modified from <4 h when the study was initiated, now eliminated as a time measure entirely).

OBJECTIVE

The purpose of this article was to determine whether TTFA is associated with inpatient mortality.

METHODS

The records of 444 consecutive patients admitted with pneumonia at a single institution were retrospectively reviewed for a correlation between TTFA and inpatient complications, including death. Statistical significance was set at p < 0.01 due to multiple comparisons.

RESULTS

Patients whose TTFA was <4 h had more complications (27% vs. 3%; p < 0.01) including death, intensive care unit admission, and intubation. These patients were judged sicker on arrival (median Emergency Severity Index 2 vs. 3; p < 0.001) and were more likely to be triaged to a critical care bed (36% vs. 5%; p < 0.001). Shortness of breath was the only presenting factor that was more frequent in the TTFA <4-h group (61% vs. 16%; p < 0.01).

CONCLUSIONS

Shorter TTFA is not associated with improved inpatient mortality. TTFA should not be considered to be a marker of quality of care but rather a reflection of patient disease severity.

Weekly E-mail reminders influence emergency physician behavior: a case study using the Joint Commission and Centers for Medicare and Medicaid Services Pneumonia Guidelines

OBJECTIVES

Improving physician compliance with evidence-based guidelines is challenging. The authors wanted to determine if weekly e-mail reminders to emergency department (ED) staff increase compliance with Joint Commission and the Centers for Medicare and Medicaid Services (CMS) community-acquired pneumonia quality measures.

METHODS

One nurse administrator reviewed records on a weekly basis for all adult patients admitted to the hospital from the ED with a working diagnosis of pneumonia. An e-mail was then sent to all ED staff indicating the percentage of patients with antibiotic timing less than 4 hours from arrival. The names of individuals who administered antibiotics in less than 1 hour were highlighted. This study compared the time to antibiotics for 11 months before and 11 months after commencing this intervention.

RESULTS

There were 281 patients in the control cohort, and 37 met exclusion criteria, leaving 244 for analysis. There were 342 patients in the intervention cohort, and 40 met exclusion criteria, leaving 302 for analysis. The median time from arrival to chest radiograph order decreased significantly from 61 to 47 minutes (p < 0.001). The median time interval from chest radiograph order to antibiotic administration did not change significantly (92 to 88 minutes, p = 0.294). The overall median time from arrival to antibiotic administration decreased significantly from 162 to 146 minutes (p = 0.018). The percentage of patients with antibiotic administration within 4 hours increased from 77.5% to 86.1% (p = 0.009).

CONCLUSIONS

Weekly e-mail reminders listing performance on antibiotic administration recommendations are associated with increased compliance with a clinical guideline.