Rising antibiotic resistance inflicts a heavy burden on healthcare, both clinically and economically. Owing to the time required to obtain culture and sensitivity test results, quite often the clinicians rely on their experience and static clinical guidelines to prescribe antibiotics. Such empirical treatment often fails to account for patient-specific attributes and changes in the antibiotic resistance patterns with time and location. The aim of this study was to analyze the patient and hospital specific features regarding their prognostic relevance to treat bacterial infections of patients in the intensive care units (ICUs). We performed a single-center retrospective cohort analysis across 25526 positive cultures recorded in MIMIC-III critical care database. We retrieved a number of clinically relevant relationships from association analysis between patient factors and bacterial strains. For instance, higher elapsed time from patient admission to sample collection for culture showed strong association with blood stream infection caused by Enterococcus faecium, Pseudomonas aeruginosa, and Staphylococcus, indicating that these infections are possibly hospital acquired. To predict antibiotic sensitivity at the level of individual patients we developed an ensemble of machine learning algorithms. The model provided superior prediction accuracy (about 87%) and area under the ROC curve (around 0.91 on an average) for the four most common sample types as compared to a number of off-the-shelf techniques. We demonstrate the predictive power of commonly recorded patient attributes in personalised prediction of antibiotic efficacy.
The aggravating loss of antibiotic activity as a result of the dispersal of resistant genes among micro-organisms has become a global health challenge and a threat to humankind. Antibiotic resistance inflicts a heavy burden on healthcare both clinically and economically, with 23,000 and 25,000 estimated annual deaths respectively in the United States and in Europe as well as increasing length of stay and morbidity. Reports suggest a projected annual death toll touching ten million worldwide by 2050 . In the United States alone, the annual cost associated with antimicrobial resistance has been estimated to be $55 billion. Intensive Care Units (ICUs) in hospitals are often considered to be the epicenter of development, acceleration, and proliferation of drug-resistant microorganisms. Critically ill patients in ICU are particularly vulnerable to infections due to their exposure to multiple invasive procedures. These include mechanical ventilation, tracheal intubation, vascular access etc. which, in turn, leads to compromised defense mechanism of anatomical barriers, impairment of protective mechanisms such as cough reflex by sedative drugs, and the frequent impairment of the immune response induced by trauma, surgery, and sepsis. According to a dated multi-center study (conducted in 1992) in Europe, around 20% to 30% of ICU admissions had reported an incidence of nosocomial infection. A more recent study (conducted in 2007) involving 1265 ICUs from 75 countries reported the presence of hospital-acquired infections in about 50% of ICU patients. In today’s world of rapidly growing antimicrobial resistance (AMR), optimal antibiotic prescription is crucial in critical care settings. Frequently used broad-spectrum antibiotics are among the primary drivers of rising AMR. Reports suggest that 30% to 60% of the antibiotics prescribed in ICUs are unnecessary, inappropriate, or suboptimal. Epidemiological studies have demonstrated a direct relationship between antibiotic consumption and the emergence/propagation of several resistant strains in ICUs. Nosocomial infections, caused by multi-drug resistant (MDR) organisms, are more prevalent in ICUs as compared to other departments. Infection caused by MDR organisms often result in much worse clinical outcomes compared to their susceptible counterparts. Such outcomes have also been linked with delay in the administration of right antibiotics. In the presence of clinical symptoms, detection of the pathogen via culture remains the gold standard for diagnosing the majority of the bacterial infections including urinary tract infection (UTI) and bloodstream infection (BSI). Culture and Sensitivity (C & S) report provides definitive evidence of a particular infection in a subjected specimen. C & S reports also cite antimicrobial susceptibility of the individually tested drugs through Minimum Inhibitory Concentration (MIC) values. Generation of a C & S report typically takes around 24 to 72 hours. In its absence, physicians rely on their perception about the clinical presentation of the specific cases and other available clinical guidelines. Such approaches could potentially disregard patient-specific attributes and the temporal changes in the antimicrobial resistance patterns. Appropriate choice of empiric antibiotic must balance out the objective to minimize the prescription of broad-spectrum antimicrobial agents and give a broader spectrum of coverage across various bacterial strains. It has been shown that inappropriate empiric therapy is associated with poorer outcomes with longer length of stay, increased health-care costs, higher morbidity, and mortality. Numerous studies have suggested appraisal of pathogen etiology in a localized setting as the ideal basis of empiric therapy. We performed a retrospective study to correlate antibiotic resistance to a broad range of patient-specific factors such as gender, comorbidities, site of infection, the events of past hospitalization, and previous antibiotic usage. These factors show significant non-monotonic associations with the efficacy of the antibiotics. We used patient information and culture data of 11496 patients from the Medical Information Mart for Intensive Care III (MIMIC-III, data collected between 2001 and 2012) critical care database. We first analyzed the bacterial prevalence and susceptibility patterns across bacteria-antibiotic pairs. Further, we presented a statistical approach unraveling the complex landscape of association between various patient-related features and the bacterial species. This analysis provided us with a comprehensive set of clinically relevant relationships. Finally, we explored the potency of various concerned patient factors to predict the susceptibility of bacteria to specific antibiotics. To this end, we developed an ensemble of machine learning models for personalized antibiotic susceptibility prediction that yielded a high overall accuracy, compared to some of the existing best practice methods.