By admin 
The data emerging from healthcare facilities across the Commonwealth suggests a crisis of unprecedented proportions. Sepsis, the body’s extreme and life-threatening response to an infection, has long been a leading cause of death in American hospitals. However, recent year-over-year statistics show a sharp, jagged climb in the incidence of sepsis diagnoses that seems to outpace the aging population and the prevalence of underlying comorbidities. While public health officials initially feared a new, more virulent strain of antibiotic-resistant bacteria or a systemic failure in infection control, a more technical and perhaps more controversial explanation is gaining traction among healthcare analysts: the rapid integration of artificial intelligence in the medical coding and billing process.
Medical coding is the critical, if invisible, backbone of the healthcare economy. It is the process by which a physician’s clinical notes, lab results, and procedures are translated into standardized alphanumeric codes—primarily the International Classification of Diseases (ICD) system. These codes determine how much a hospital is reimbursed by insurance companies and government programs like Medicare and Medicaid. Traditionally, this was the domain of human specialists who pored over charts to ensure accuracy. However, in the last five years, hospitals across Massachusetts and the nation have pivoted toward Computer-Assisted Coding (CAC) and fully autonomous AI coding systems. These tools use Natural Language Processing (NLP) to scan electronic health records (EHR) with a speed and granularity that no human could match.
The surge in sepsis cases, experts suggest, may be a byproduct of these algorithms doing exactly what they were designed to do: maximize documentation and identify every possible billable condition. Sepsis is a particularly attractive target for AI-driven "optimization" because it is a high-weighted Diagnosis Related Group (DRG). In the complex world of hospital finance, a diagnosis of sepsis can significantly increase the reimbursement for a patient’s stay compared to a diagnosis of a simple localized infection, such as pneumonia or a urinary tract infection.
"What we are seeing is a divergence between clinical reality and administrative data," says Dr. Aris Nikas, a health policy researcher who has studied the impact of algorithmic billing. "If you look at the billing codes, it looks like we are in the midst of a sepsis epidemic. But if you look at the clinical outcomes—the number of patients actually requiring vasopressors or experiencing multi-organ failure—the trend is much flatter. The AI is finding ‘sepsis’ in the data that a clinician might simply call a ‘bad infection.’"
The crux of the issue lies in the definition of sepsis itself, which has been a subject of intense debate within the medical community for decades. Currently, there are two primary frameworks: Sepsis-2 and Sepsis-3. Sepsis-2, established in 2001, is a broader definition based on the Systemic Inflammatory Response Syndrome (SIRS) criteria, which include common symptoms like elevated heart rate, fever, or an increased white blood cell count. Sepsis-3, introduced in 2016, is much more stringent, requiring evidence of life-threatening organ dysfunction.
AI coding systems are often programmed to flag any instance where a patient’s lab results and vital signs meet the broader Sepsis-2 criteria. Because AI can monitor every heartbeat and every blood draw recorded in an EHR in real-time, it can identify a "sepsis event" the moment a patient’s heart rate ticks up and their white blood cell count fluctuates, even if the attending physician never considered the patient to be in septic shock. When the AI suggests a sepsis code, and the physician signs off on it—often during a busy shift where clicking "agree" is the path of least resistance—the hospital’s revenue increases.
This phenomenon, often referred to as "diagnostic creep" or "upcoding," is not necessarily illegal, but it creates a distorted picture of public health. In Massachusetts, where the healthcare market is dominated by high-cost academic medical centers, the financial stakes are enormous. The state has some of the highest healthcare expenditures per capita in the world, and any shift in coding practices that inflates the severity of illness can add hundreds of millions of dollars to the total cost of care.
Critics of the AI-coding surge argue that this creates a "ghost epidemic." When hospital performance is measured by sepsis mortality rates, the data can be easily manipulated. If a hospital uses AI to label thousands of mildly ill patients with a "sepsis" code, and those patients all survive, the hospital’s sepsis mortality rate appears to plummet, making them look like leaders in quality care. In reality, the patient population didn’t change; only the labels did.
"It’s a statistical shell game," notes Sarah Thompson, a former hospital administrator now working as a consultant for insurance payers. "The AI isn’t just a tool for efficiency; it’s a tool for revenue cycle management. By capturing ‘query-able’ opportunities for sepsis, hospitals are essentially mining their own data for profit. The problem is that this data is then used by researchers and policymakers to determine where to allocate resources. If the data is skewed by AI, the policy decisions will be flawed."
However, proponents of AI in healthcare offer a different perspective. They argue that sepsis has historically been underdiagnosed and that human coders frequently miss the subtle indicators of systemic inflammation. In this view, AI is not "creating" cases but is finally capturing the true burden of the disease. They point out that early intervention is the only way to survive sepsis, and if an algorithm alerts a team to a potential case earlier than a human would have noticed, lives are saved.
"The goal of AI in coding is to ensure that the medical record is a complete and accurate reflection of the patient’s clinical status," says James Sterling, a Chief Information Officer for a major Boston hospital network. "Sepsis is a complex, fast-moving condition. If our systems can help identify the severity of an illness more accurately, that allows us to allocate the right resources to that patient and ensures the hospital is fairly compensated for the high-intensity care we provide."
The tension between these two views is now reaching a boiling point. The Centers for Medicare & Medicaid Services (CMS) and private insurers are beginning to take notice of the "sepsis jump." Audits are becoming more frequent, and there is growing pressure to standardize which definition of sepsis is used for billing. In 2025, several major insurers began denying claims for sepsis that did not meet the stricter Sepsis-3 clinical criteria, leading to a legal standoff between payers and providers.
In Massachusetts, the Health Policy Commission (HPC) has expressed concern about the rising costs associated with these trends. While the state prides itself on being a hub for technological innovation in medicine, there is a growing recognition that technology must be tempered by oversight. The fear is that if AI is left to optimize billing without clinical guardrails, the result will be a permanent inflation of healthcare costs that provides no actual benefit to patient health.
Beyond the financial and statistical implications, there is a clinical risk to AI-driven sepsis coding. If every infection is coded as sepsis, there is a risk of "alert fatigue" among clinicians. Furthermore, if the medical record is cluttered with high-severity diagnoses for billing purposes, it can obscure the actual clinical priorities for the patient. A patient who is treated for "sepsis" based on a coding algorithm might receive aggressive fluid resuscitation and broad-spectrum antibiotics that they don’t actually need, leading to complications like fluid overload or the development of C. diff.
As we look toward the late 2020s, the intersection of AI and healthcare finance remains one of the most volatile areas of the industry. The Massachusetts experience serves as a microcosm for a global challenge: how to utilize the power of machine learning to improve efficiency without sacrificing the integrity of medical data.
The "alarming increase" in sepsis cases in Massachusetts may indeed be a sign of a changing world, but perhaps not in the way we think. It is not necessarily a sign of a new pathogen or a failing of the human body, but rather a sign of the incredible, sometimes unintended power of the algorithms we have built to track us. Until the medical community can agree on a single, clinical definition of sepsis that AI must adhere to, the numbers will likely continue to climb, leaving experts to wonder where the biology ends and the billing begins. The jump in cases is a reminder that in the age of AI, data is never just a neutral reflection of reality—it is a product of the tools we use to measure it.