If You Can Fill a Spreadsheet, You Can Build an Assessment

a month ago

Illustration

The Context

Healthcare assessments are everywhere—PCOS risk calculators, breast cancer screening tools, fertility questionnaires. At Proactive For Her, our medical team was constantly requesting new assessment tools to engage patients and provide personalized health insights. "Can we build a cervical cancer risk assessment?" "We need an FSFI questionnaire for sexual health." "What about a hormonal imbalance screener?"

Each request followed the same painful pattern: medical team creates requirements, developers spend hours coding custom logic, back-and-forth iterations to get the scoring right, and by the time it's live, we've lost critical time. We were treating every assessment like a software project when what we really needed was a content management system for healthcare logic.

The breaking point came when our gynecologist wanted to launch three new risk assessments in a single month. The traditional approach would have required 6+ weeks of developer time, not counting the inevitable revisions when medical scoring needed adjustment.

The Problem

Our assessment creation process was fundamentally broken:

  1. Developer bottleneck for medical content: Every new assessment required custom JavaScript logic. Changing a risk factor weight meant deploying new code. Medical professionals couldn't iterate on their own expertise—they had to wait for engineering sprints.
  2. No complex decision support: Healthcare assessments need sophisticated logic—vetoes (immediate disqualification), breakpoints (threshold-based outcomes), and postprocessors (conditional outcome adjustments). Our system couldn't handle "if HPV vaccinated AND recent Pap test, downgrade from high risk to medium risk."

The complexity was particularly acute for women's health, where risk factors interact in nuanced ways. Age, reproductive history, lifestyle factors, and screening history all influence outcomes, often with conditional logic that changes based on other responses.

The Solution

The Matrix: Mapping Medical Knowledge to Digital Logic

Healthcare professionals think in decision matrices—"if patient is 50+, that's high risk; if 30-49, medium risk; if under 30, low risk." We built a system that let them express this knowledge directly in spreadsheet-like tables.

OutcomeName/QuestionName  | High Risk | Medium Risk | Low Risk
CVAge                     | 50+       | 30-49       | <30
CVSmoking                 | Y         | UsedTo      | N
CVSexPart                 | tom       | ott         | zero

Medical teams could modify risk factors instantly. Discover new research showing that 45+ should be high risk instead of 50+? Change one cell in the CMS. No code deployment, no developer tickets, no delays.

Vetoes: Absolute Certainties

Some medical conditions create absolute certainties. A person assigned male at birth cannot get cervical cancer. Someone who's had a complete hysterectomy with cervix removal has zero cervical cancer risk. These aren't risk factors—they're medical impossibilities.

Vetoes run first, before any other scoring. They're binary decisions that override everything else:

  • SexChoice = Male → Outcome: Undetermined (assessment not applicable)
  • Hysterectomy = Yes → Outcome: Undetermined (anatomically impossible)
  • PreviousCancerDiagnosis = Yes → Outcome: High Risk (requires immediate medical attention)

Vetoes ensured that our assessments never provided medically impossible or dangerous advice.

Breakpoints: Factor Accumulation

Healthcare risk assessment isn't just about individual factors—it's about factor accumulation. One high-risk factor might be manageable. Three high-risk factors demand immediate attention. Breakpoints encoded this medical reasoning.

Outcome    | Priority | Breakpoint
High Risk  | 3        | 2         (2+ high-risk factors = high risk)
Medium Risk| 2        | 2         (2+ medium-risk factors = medium risk)  
Low Risk   | 1        | 1         (1+ low-risk factor = low risk)

Priority ordering mattered critically. If someone had 2 high-risk factors, 7 medium-risk factors, and 9 low-risk factors, they got "High Risk"—not medium or low. The algorithm checked breakpoints in priority order, assigning the first threshold met.

Postprocessors: Conditional Logic

Healthcare is full of conditional logic that simple scoring can't capture. Someone might score as "High Risk" based on age and lifestyle factors, but if they're up-to-date on screening and vaccinated, their actual risk is much lower. Postprocessors handled these nuanced medical interactions.

IF initial_outcome = "High Risk" 
AND HPV_vaccination = "Yes" 
AND pap_test_recent = "Yes"
THEN final_outcome = "Medium Risk"

This encoded real medical guidelines. HPV vaccination plus recent screening genuinely reduces cervical cancer risk, even for people with other risk factors. The postprocessor system let medical teams encode these evidence-based modifications.

The Decision Engine: Orchestrating Medical Logic

The genius was in how these systems worked together, following medical reasoning patterns:

  1. Vetoes First: Check for absolute medical disqualifiers
  2. Matrix Mapping: Categorize all responses into risk buckets
  3. Breakpoint Analysis: Count factors and assign initial outcome
  4. Postprocessor Refinement: Apply conditional modifications

This order matched clinical decision-making. Doctors first rule out impossibilities, then assess risk factors, then apply evidence-based modifications. Our algorithm encoded medical reasoning patterns.

The Impact

  1. Assessment creation speed: New assessments went from 6+ weeks to 1-2 days. Medical teams could create and iterate on risk calculators independently, without developer bottlenecks.
  2. Medical sophistication: Successfully implemented complex healthcare algorithms that actually matched clinical reasoning patterns. Vetoes prevented impossible outcomes, breakpoints encoded factor accumulation logic, and postprocessors captured evidence-based modifications.
  3. Clinical accuracy: The four-layer decision system (vetoes → matrix → breakpoints → postprocessors) matched how medical professionals actually think about risk assessment.
  4. Medical team autonomy: Gynecologists, endocrinologists, and other specialists could build sophisticated assessment tools that reflected their clinical expertise, without waiting for developers to interpret their requirements.

When doctors can encode their expertise directly—vetoes, breakpoints, conditional logic, factor interactions—the resulting tools are medically accurate and clinically relevant.