This document explains the pre-defined, heuristic weights used in our system.
This document explains the pre-defined, heuristic weights used in our system. These are numerical values based on established medical guidelines and expert opinion, designed to quantify the relative importance of different risk factors.
Purpose: Instead of asking the machine learning model to learn everything from scratch, we are "pre-loading" it with clinical domain knowledge. This creates powerful, high-level features (composite risk scores) that make the model more accurate, more stable with limited data, and significantly more interpretable.
This score aggregates a patient's current health conditions. The weights reflect the typical severity and impact of each condition on pregnancy outcomes.
| Risk Factor | Weight | Clinical Rationale & Example |
|---|---|---|
has_diabetes, has_heart_disease |
3.0 | Rationale: These are major systemic diseases that significantly increase the risk of severe maternal and fetal complications. They are given the highest weight in this category. Example: A patient with pre-existing diabetes is assigned a higher baseline risk than a patient with only a thyroid condition. |
hb_severity, has_hypertension, has_kidney_disease |
2.5 | Rationale: Severe anemia, high blood pressure, and kidney disease are also very serious risk factors that can lead to poor outcomes like pre-eclampsia or fetal growth restriction. Example: A patient with a hemoglobin level of 8 g/dL (Moderate Anemia) gets a higher risk score contribution than a patient with a BMI of 31 (Obese). |
has_autoimmune, bp_risk_score |
2.0 | Rationale: Autoimmune conditions and elevated blood pressure represent a substantial, but often more manageable, risk compared to the conditions above. |
bmi_risk_score, has_thyroid |
1.5 | Rationale: While important, abnormal BMI and thyroid issues are common and often well-managed during pregnancy. They contribute to overall risk but are weighted lower than acute systemic diseases. |
This is a critical score, as past obstetric performance is one of the strongest predictors of future outcomes. The weights are designed to capture the severity and, most importantly, the pattern of previous losses.
| Risk Factor | Weight | Clinical Rationale & Example |
|---|---|---|
all_previous_lost |
5.0 | Rationale: This is the most severe historical indicator. A history of zero successful live births despite multiple pregnancies suggests a significant underlying issue. Example: A patient who is G3 P0 (3 pregnancies, 0 live births) receives the highest penalty, signaling a critical history. |
has_recurrent_loss (3+), has_stillbirth, has_child_death, high_loss_ratio |
4.0 | Rationale: These factors indicate an established pattern of significant loss. A stillbirth or multiple abortions are clinically more alarming than a single early miscarriage. Example: A patient with a history of one stillbirth (S=1) gets a higher risk score than a patient with a single first-trimester abortion (A=1). |
has_multiple_abortions (2+), bad_obstetric_history |
3.0 | Rationale: A history of two or more losses is a clear step-up in risk from a single, isolated loss and is a standard clinical marker for further investigation. |
has_abortion |
2.0 | Rationale: A single past abortion is a known risk factor, but it is weighted as the lowest among the loss-related events as it can often be an isolated incident. |
These scores quantify risks from physical uterine changes and biological age-related factors.
| Risk Factor | Weight | Clinical Rationale & Example |
|---|---|---|
multiple_lscs |
3.0 | Rationale: The risk of complications like uterine rupture or placenta accreta increases with each successive C-section. This weight is double that of a single C-section. |
has_lscs |
1.5 | Rationale: A single previous C-section carries risk for future pregnancies and is an important factor to consider. |
teenage_pregnancy, advanced_maternal_age |
3.0 | Rationale: This reflects the well-known "U-shaped" risk curve of maternal age. Both very young and advanced-age mothers face higher risks of complications. |
Finally, the sub-scores are combined into a single, holistic `total_risk_score`. The weights here represent a high-level clinical hypothesis about which *categories* of risk are most predictive overall.
| Risk Category | Weight | Clinical Rationale |
|---|---|---|
| Pregnancy History Risk | 0.4 (40%) | A patient's own obstetric history is often the single most powerful predictor of future pregnancy outcomes. |
| Medical Risk | 0.3 (30%) | Current maternal health status is the next most critical factor. |
| Age-Related Risk | 0.2 (20%) | Maternal age is a significant, independent risk factor. |
| Surgical Risk | 0.1 (10%) | While important, surgical history is considered a less impactful predictor than the other categories in the general high-risk population. |
The current system of heuristic weights is robust and well-justified. However, a key role of an expert is to also identify the current limitations and define a path forward. The "missing" elements are not oversights, but rather opportunities for the next version of the system, contingent on acquiring more detailed data.
The model's accuracy could be further improved by incorporating more specific clinical data, as noted in the `DataCollectionStrategy`. Key missing data includes:
The current model provides a highly valuable static risk score, typically calculated at the first prenatal visit. A future enhancement (Version 4.0) would be to create a dynamic risk model. This would involve:
The current heuristic scores are largely additive. While the machine learning model can learn some non-linear relationships, a future version could explicitly engineer features that model complex interactions. For example:
diabetes_x_hypertension_risk: A feature that assigns a much higher risk if both diabetes and hypertension are present, as their combined effect is often synergistic (more than the sum of their parts).By presenting these as a clear roadmap, we demonstrate a deep understanding of the problem space and a commitment to continuous improvement.