The role of artificial intelligence in healthcare continues to evolve. Key stakeholders in the healthcare system have graduated from the realm of pure skepticism and now invite AI to prove itself in real-time. Within a given hospital, AI algorithms may be deployed in a variety of subspecialty domains, including cardiology, radiology and oncology. As the clinical arena adopts new AI-driven technologies, metrics to assess their inherent value must be developed. Vendors, physicians and hospitals must ask themselves how they can best quantify the impact of AI once deployed in their clinical arena. Specifically, it is vitally important for early adopters of artificial intelligence to harness data and metrics that depict how the technology impacts the goals of their enterprise. A unique feature of diagnostic radiology is its direct impact on nearly every specialty in the hospital. Consequently, when integrating AI-based solutions into the field of radiology, it is prudent to identify key clinical metrics that may inform us of the most pressing question: What is the impact of AI in radiology?
According to the Agency for Healthcare Research and Quality, the national average for a hospital stay is 4.5 days, at an average cost of $10,400 per day. The amount of time a patient remains in the hospital for a medical intervention is directly correlated with the quality of care offered at a hospital. Furthermore, studies have demonstrated that prolonged lengths of stay are associated with an increased rate of hospital-acquired infections and medication side effects. From a clinical perspective, evidence suggests that a reduced LoS may decrease the incidence of infections and mortality. Importantly, a 2011 study demonstrated that medical complication rates were higher in patient groups with a prolonged length of stay when compared to those with shorter stays. As a result, clinicians are driven to reduce inpatient length of stay, as it improves clinical outcomes. Beyond the positive impact a reduced LoS may have on patient outcomes is the profound impact this key metric has on cost containment efforts. Medical management has been keenly focused on LoS reduction for over 3 decades, knowing full well that with the right team and expertise, a reduced LoS would not compromise quality of care. As a general rule, decreasing lengths of stay impacts hospitals and patients in the following domains:
Clinical: The longer an individual occupies a bed in a hospital, the more likely they are to become infected with hospital-acquired diseases. Moreover, an extended stay demands clinical attention and resources that may be better leveraged elsewhere.
Rewind a decade and you will find that most healthcare systems were generally opposed to the adoption of AI in their workflow. At the time, the verdict was out as to the value of leveraging AI into the healthcare system. With over a decade of trial and error behind it, the judgment on AI technologies is clear: AI is here to stay. It has proven itself to be invaluable from an economic, clinical, and operational perspective. Now, both community and academic healthcare systems are scrutinizing a myriad of AI-based products to determine how their deployment best fits their hospital’s mission and vision towards the delivery of quality care. Inculcating AI-based tools in the reading room as a pixel analyzer towards the detection of acute pathologies has proven its most immediate effect rather quickly: radiologists who use AI have a higher sensitivity and specificity in the detection of a host of pathologies. Since the reading room is the hub of the hospital controlling a wide spectrum of clinical decisions, it is not surprising that implementing tools at the epicenter of the hospital – the reading room – directly impacts a coveted metric in the healthcare ecosystem: LoS.
A recent study from Cedars-Sinai Medical Center highlighted the impact of implementing Aidoc to flag suspected pulmonary emboli and intracranial hemorrhage. When comparing the average LoS for patients with either PE or ICH pre-AI and post-AI activation, the data demonstrated a reduction in LoS in the post-AI cohort for both disease states. Specifically, among patients diagnosed with ICH or PE when comparing the pre-AI and post-AI time periods, a 1.30 days reduction in LoS was observed in patients diagnosed with ICH, and a 2.07 days reduction in LoS was observed in patients diagnosed with PE. The authors state that, “these changes in LOS in the ICH and PE cohorts from pre-AI to post-AI periods suggest a change due to the triage software implementation.” Additionally, researchers from Yale Medical Center presented data depicting the power of AI within the emergency department as it pertains to LoS. In analyzing over 25,000 ICH positive CT cases, the authors showed that after the utilization of Aidoc the turnaround time dropped from 53 minutes to 46 minutes. More importantly, the LoS for patients within the emergency department decreased from 567 minutes to 508 minutes. Whether in the ED or on the hospital floors, it is abundantly clear that amplifying detection speed and prioritization strengths of the reading room translates into meaningful minutes saved throughout both the inpatient and ED setting. The efficiency gains reaped through AI is not confined to the realm of economics and cost containment, but also has practical implications for health outcomes.
Technological innovation of all forms in healthcare boasts its ability to get the right patient the right treatment at the right time, but how exactly is this accomplished? In what way can decision makers within healthcare measure a technology’s effectiveness at achieving this goal? Studies have shown a direct correlation between time to treatment and patient outcomes. A two hour reduction in the time from hospital arrival to the start of anticoagulation therapy has been shown to significantly increase PE patient survivors. For cases of intracranial hemorrhage, hematoma expansion occurs typically in the first few hours after bleeding starts, and hematoma volume is an important early predictor of deterioration. Therefore, promptly identifying patients with ICH, and getting them the appropriate treatment they need, may be the underlying driver of decreased LoS associated with AI-based detection of ICH.
Healthcare as an ecosystem is constantly bombarded with novelty and innovation. Whether it be artificial intelligence in radiology, genomic sequencing in oncology, or robotic assisted devices in the operating room, hospitals must utilize a framework by which to judge the value and impact of new technologies. The deployment of AI as a care coordination utility, from the reading room to the patient bedside, proves itself to have a direct impact on patient care well beyond the tangible reductions it may offer in lengths of stay.