In March 2026, two oncologists were arrested in Cordoba, Argentina, charged with issuing false prescriptions against the provincial health plan. The investigation, which began in 2021, had gathered evidence for five years before prosecutors obtained arrest warrants. During that entire period, the physicians continued prescribing, pharmacies continued dispensing, and the health plan continued paying.
This case is not an isolated incident. Provider fraud — physicians, pharmacists, and other health professionals who exploit their position of trust to generate fraudulent claims — is a documented problem across Latin America. What makes it particularly difficult to detect is that these professionals have legitimate access to the system. Their prescriptions are formally valid. Their credentials are current. Their patients exist.
The time problem
Five years. That is the timeline of the Cordoba case, from the start of the investigation to the arrests. This is not an atypical timeline for provider fraud investigations. Manual medical fraud investigations require documentary evidence collection, interviews, prescription pattern analysis, and construction of a legal case that can withstand judicial scrutiny.
The cost of that slowness is twofold. First, the direct financial cost: fraudulent dispensations continue throughout the investigation period. Second, the opportunity cost: investigation resources dedicated to a single case for years are unavailable to investigate others.
The question this case raises is whether there is a way to identify anomalous prescribing patterns in weeks rather than years. The technological answer is yes.
What signals anomalous prescriber behavior generates
Prescriber fraud generates detectable signals when claims data is analyzed as a whole. Inspector AI monitors prescriber behavior across multiple dimensions, each of which can generate independent signals that, combined, point with high precision to providers requiring investigation.
Prescribing concentration. A physician whose prescribing activity is disproportionately concentrated in a specific medication or therapeutic group generates a statistical signal. If an oncologist prescribes a volume of a specific treatment that exceeds the mean of other oncologists in the same region by multiple standard deviations, that is a detectable anomaly.
Polypharmacy patterns. Prescribers whose patients systematically receive unusual medication combinations generate polypharmacy signals. The system evaluates these combinations against established clinical protocols, and when 80% of signals are correctly exempted by clinical logic (as occurs in oncology, HIV, and cardio-metabolic conditions), the remaining signals have significantly higher predictive value.
Patient overlap. When a prescriber shares an unusually high percentage of patients with another prescriber, especially if both prescribers are generating high dispensation volumes, the signal suggests coordination. Each individual prescription may be legitimate in form, but the pattern of relationships between prescribers is anomalous.
Prescribing frequency. The number of prescriptions issued per unit of time is a fundamental signal. A physician issuing more prescriptions than the 99th percentile of their specialty generates a signal that merits investigation, regardless of whether each individual prescription appears correct.
Diagnostic consistency. When a physician's prescriptions show a systematic disconnect between recorded diagnoses and prescribed medications — what Inspector AI detects as clinical mismatch — the signal is particularly strong. An oncologist prescribing oncological treatments to patients with no recorded oncological diagnosis generates a clear and immediate signal.
From signal to investigation
It is important to be precise about what automated detection does and does not do. It does not replace investigation. It does not determine guilt. What it does is radically reduce the time between the appearance of an anomalous pattern and its identification for investigation.
In the manual model, someone has to notice that something is wrong. That can take months or years, or it may never happen if the transaction volume is large enough and each individual transaction stays within normal parameters. In the automated model, signals are generated as soon as the pattern establishes itself in the data, which can be a matter of weeks.
This does not mean Inspector AI would have detected the specific case of the Cordoba oncologists. We do not know the operational details of the scheme or the data structure of the provincial health plan. What we do state is that prescriber anomaly detection is a core capability of the platform, and that five-year investigation timelines are a symptom of manual-only processes.
The Latin American context
Provider fraud in Latin America has characteristics that make it particularly challenging. Health system fragmentation means a physician can prescribe to multiple payers without any single one having complete visibility into their behavior. Lack of interoperability between information systems prevents data correlation across institutions. And the scarcity of health professionals in many regions creates reluctance to investigate active physicians for fear of reducing service supply.
These limitations do not eliminate the possibility of detection. They reduce it. An individual payer can analyze the behavior of its prescribers within its own database and detect significant anomalies. It does not need system-wide visibility to identify that a prescriber shows patterns that deviate radically from their peers.
Beyond the individual case
The value of systematic prescriber anomaly detection goes beyond identifying fraud cases. The same analyses that detect fraudulent prescribing also detect poor prescribing: physicians who systematically prescribe branded drugs when generics are available, who generate unnecessary early refills, or who prescribe combinations with interaction risk.
This means that investment in prescriber anomaly detection has returns that go beyond fraud prevention. It is an investment in prescribing quality, adherence to clinical protocols, and pharmaceutical efficiency.
The first step
For insurers that suspect they have a provider fraud problem but lack the tools to quantify it, the first step is analyzing existing data. The patterns are in the data; what is missing is the capability to see them.
To request a free analysis of your claims data, visit our contact page.