Why incident diagnosis are used

Why Incident Diagnoses Are Used

What “Incident Diagnoses” Means

In this project, TreeScan is run on incident diagnoses. An incident diagnosis is a diagnosis that is treated as new for a patient at the time it appears in the study period.

This means that if a patient has multiple ED visits with the same diagnosis across time, only the first occurrence (within the relevant definition window) is counted as incident, and subsequent repeat occurrences are not counted as new cases for the purpose of signal detection.

---

Why Incident Diagnoses Matter for Syndromic Surveillance

Emergency department diagnosis data include both:

• new acute events (e.g., acute infection, injury, poisoning)
• chronic conditions (e.g., diabetes, hypertension, asthma)
• repeat visits for ongoing issues (e.g., follow-up care, persistent symptoms)

If we counted all diagnoses on all visits, then signals could be driven by:

• repeated visits by the same patients
• ongoing chronic disease burden
• differences in care-seeking behavior rather than new events

This is not ideal when the goal is to detect emerging health events.

Using incident diagnoses reduces the influence of repeat encounters and focuses detection on new occurrences of diagnoses.

---

What Problem Incident Diagnoses Prevent

Without incident diagnosis logic, a signal could arise simply because:

• the same individuals repeatedly return for the same condition
• a hospital changes follow-up scheduling behavior
• a subset of patients has frequent revisits due to social or healthcare access factors

These patterns may be important for healthcare operations, but they are not necessarily signals of new public health events.

Incident diagnoses therefore help align TreeScan signals with the concept of:

“Are we seeing more new diagnoses than expected?”

rather than:

“Are we seeing more total visits with a diagnosis than expected?”

---

How Incident Diagnoses Are Determined

To decide whether a diagnosis is incident, the system must determine whether the same patient had that diagnosis previously.

This requires:

• a patient identifier (e.g., unique patient ID)
• historical ED visit data prior to the study period
• logic that checks whether the diagnosis appears for that patient in the lookback window

If a patient had the diagnosis before, then the diagnosis is not treated as incident when it appears again during the study period.

---

Why This Requires a Lookback Period

Because incident diagnoses require checking prior history, the analysis must include data from before the study period.

In this project, we use a one-year lookback.

This provides a practical window for determining whether a diagnosis is plausibly “new” for surveillance purposes, while balancing feasibility and data availability.

---

Why This Creates the 15-Month Data Pull

The analysis design uses:

• 90 days as the study period (the window in which clusters are evaluated)
• 1 year of lookback prior to the start of the study period to determine incident diagnoses

Therefore, to run one analysis window, the data needed include:

• 12 months of lookback data
• 3 months (90 days) of study data

This totals approximately 15 months of ED visit data.

---

Practical Interpretation

Using incident diagnoses is a design choice intended to make TreeScan outputs:

• more interpretable for public health surveillance
• less sensitive to repeat visits by the same individuals
• more reflective of emerging changes in new diagnoses rather than ongoing healthcare utilization patterns

This is especially important in operational settings where signals must be reviewed routinely and where false alerts caused by repeat utilization patterns would add unnecessary burden.