AI, Algorithms and Clinical Judgment

Assistive AI algorithms in medical procedures are designed to enhance decision-making, improve accuracy, and increase efficiency in healthcare settings. Many different types exist in new technology. For example, Predictive Analytics. These algorithms use historical patient data to predict specific outcomes, such as alterations in complications. Machine learning models can analyze factors in an “if-to” pathway. The unit will prompt “if” a predictive path and pre-programmed models actions into effect. AI tools facilitate remote and telemedicine patient monitoring by analyzing data from wearable devices, such as ecg’s. The algorithms can detect anomalies in waveforms, analyze the regularity, vital signs and alert healthcare providers or a specific change.

These assistive AI algorithms are transforming medical procedures by enhancing accuracy, reducing human error, however nothing can replace human clinical judgement and experience. Computers should augment human intelligence rather than replace it, right?

Today, many people perceive advanced computers as intelligent due to their ability to learn and act based on the data they process. However, this form of intelligence is fundamentally different from the intelligence inherent in humans. Human abilities are far more extensive; while computers may react based on existing algorithmic data, humans possess the unique capacities to imagine, anticipate, feel, and analyze evolving situations allowing them to navigate from immediate concerns to long-term perspectives. In our case, we are presented with a patient in his early 30s with a complex medical history that includes obesity and diabetes. During our audit of our central venous access device insertions (CVAD), our team identified an issue with the patient's ECG, prompting a consultation. To summarize the history of present illness (HPI), long-term antibiotic therapy necessitated the placement of a peripherally inserted central catheter (PICC). However, the native ECG warrants attention due to the patient's underlying disease process. Given the significant history of insulin-dependent diabetes mellitus (IDDM), hypertension (HTN), and obesity, the patient may have undiagnosed cardiomyopathy.

Evidence of bi-ventricular hypertrophy is apparent in all leads, particularly in lead Il, where the native tracing displays a large bidirectional QRS complex. It is important to note that the ECG criteria for identifying right or left ventricular hypertrophy have low sensitivity (approximately 50%). meaning that about half of patients with ventricular hypertrophy may go undetected using these criteria. However, they exhibit high specificity (over 90%), indicating that when the criteria are met, it is very likely that ventricular hypertrophy is present.

General features of left ventricular hypertrophy on ECG include elevated QRS voltage criteria (i.e., tall R-waves in left ventricular leads, deep S-waves in right ventricular leads), a left ventricular strain pattern, or ST-T changes oriented opposite to the QRS direction. Additionally, there is evidence for left atrial enlargement. The pathophysiology in this scenario is evident in the available data. The tools we utilize to guide our practice serve as just that—a guide. Knowledge, experience, and a thorough examination of the history of present illness enable us to accurately interpret this interpreted ECG. The machine relies on algorithmic interpretation to detect the augmented wave of the atrial tissue. However, given the maximal height of the ventricular depolarization, a "go ahead" signal-whether in green, blue, or another symbolic icon—is indicated.

Considering that the native ECG suggested an underlying pathology, it could have potentially led to a significant infusion complication. The line was subsequently replaced to ensure the patient's safety.