INF ECTION P R EVENTION
to which patients have the warning signs of a particular infection. The system can then notify clinicians and provide an early call to action, avoiding delays in treatment and improving outcomes.
Before we talk about some examples of AI solutions successfully tackling infections, we need to consider some of the challenges AI has to overcome. For example, it is easy to say it analyses data from the EMRs of multiple hospitals – but how do you do that if each EMR is different? We all know that the EMR is the nucleus – it is where all data is stored. Everything that happens with a patient is recorded there. Although most modern hospitals use EMRs, they might have different systems, vendors and standards, making them incompatible and the data non- transferrable. This basic problem is where AI can help, as machine learning can dig into different systems and get a consistent view across them. If one hospital uses one set of acronyms, another may be completely different. And then there are the clinicians’ narratives, often recorded in inconsistent fields, and written in the garbled notes medics are notorious for.
AI can make sense of all this, using deep learning to decipher the inconsistencies in EMR mappings, and then match them to one standard. It can also use natural language processing (NLP) to identify the important information in clinicians’ written notes and ingest it as a consistent and comparable set of data.
Multiply this on a huge scale across millions of records and you can truly harness the power of the data sitting inside EMRs. In the case of Wolters Kluwer Health’s Sentri7 and POC Advisor tools, it is possible to apply AI machine learning across the EMRs of 800 hospitals, analysing the data of 25 million patient lives.
Early warning for sepsis
One of the earliest successes we had in the application of AI was in the management of sepsis, which has historically been one of the toughest infections for hospitals to tackle. In the UK, the Sepsis Trust estimates there are around 250,000 cases of sepsis per year, resulting in nearly 50,000 deaths. In the US, at least 1.7 million adults
develop sepsis every year, resulting in nearly 270,000 deaths. Globally, it is estimated that sepsis kills 11 million people per year – more than deaths from cancer. Despite the overwhelming human and financial tolls of sepsis, efforts to improve outcomes are frustrated by the fact that sepsis mimics common illnesses, making it hard to recognise. Sepsis is clinically suspected when a patient with an infection presents with two or more criteria for systemic inflammatory response syndrome (SIRS). However, most patients with SIRS symptoms don’t have sepsis. Hundreds of various comorbid conditions and medications can mimic SIRS in different patients. This difficulty in early identification can cause fatal delays in treatment, increasing the risk of death by nearly eight percent every hour. Change management procedures have proven to be effective, such as sepsis education and screening protocols, but these can take a lot of time and be a drain on resources. Electronic surveillance systems, linked to Clinical Decision Support tools, have been widely used, but they often lack sufficient sensitivity for sepsis detection, generating too many false-positive alerts. This results in ‘alert fatigue’ and ultimately, rejection of automated alerts altogether. To address the problem of alert fatigue, our team spent four years studying the variables that led to missed cases or false positives and wrote hundreds of rules to improve the accuracy of sepsis alerts. Using an electronic surveillance system,
By analysing multiple variables and connecting the dots across data taken from EMRs, AI can provide an early warning of the patients most at risk of infection. This allows clinicians to proactively prevent infections before they start to spread.
MAY 2021
patient data was aggregated into a cloud platform. Once in the cloud, the system used an algorithm to run the data against our sepsis rules to account for acute comorbid conditions, chronic diseases, medications, and other factors specific to each patient. Lab and vital sign values and clinical parameters were also adjusted. If positive results of sepsis were found, we used POC Advisor to send alerts to clinicians to ensure rapid treatment. These alerts can be accessed from a variety of devices. After a 10-month study at Huntsville Hospital in Alabama, results revealed a 53% reduction in sepsis mortality, an alert accuracy of 95% sensitivity, and 82% specificity. Furthermore, sepsis-related 30- day readmissions dropped by 30%. These outcomes translate to Huntsville saving approximately 200 lives annually.2
Progress in the fight against CDIs C.difficile (C.diff) is another of the most dangerous and troublesome infections that AI tools are helping to tackle. With symptoms ranging from diarrhoea to life-threatening inflammation of the colon, C.diff is one of the most common microbial causes of healthcare-associated infections. While significant progress has been made to reduce the number of infections and deaths caused by C.diff, the number of overall people experiencing antibiotic resistance to infections like this remains high. In the US, C.diff infects around 500,000 patients per year, and kills nearly 30,000. In the UK, it kills around 1,500 people per year. C.diff infection (CDI) is especially difficult, and expensive, to treat because it tends to recur over time, and increases the length of stay in hospital and treatment costs. In fact, the CDC estimates that C.diff is responsible for $4.8 billion in costs in the US each year. This can be made worse by financial penalties, targeting hospitals that do not proactively focus on preventing C.diff and other HCAIs.
AI technology is helping hospitals get a
WWW.CLINICALSERVICESJOURNAL.COM l 63
▲
©nimito -
stock.adobe.com
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62 |
Page 63 |
Page 64 |
Page 65 |
Page 66 |
Page 67 |
Page 68 |
Page 69 |
Page 70 |
Page 71 |
Page 72 |
Page 73 |
Page 74 |
Page 75 |
Page 76 |
Page 77 |
Page 78 |
Page 79 |
Page 80
