Calculator Input Panel
Enter a starting probability, select risk factors, then set each network node as positive, negative, or unknown. You can also edit sensitivity and specificity values for local assumptions.
Example Data Table
| Scenario | Prior | Risk Inputs | Evidence Inputs | Model Note |
|---|---|---|---|---|
| Routine screening | 2% | No contact, normal immune status | No symptoms, unknown tests | Mostly prior driven |
| Contact investigation | 5% | Household contact, medium setting risk | Cough positive, fever positive, imaging unknown | Risk and symptoms increase odds |
| Testing review | 8% | Confirmed close exposure | IGRA positive, chest imaging positive | Diagnostic evidence dominates |
| Rule-down review | 10% | Moderate immune risk | Sputum negative, imaging negative | Negative tests reduce odds |
Formula Used
The calculator uses a Bayesian odds update. This is similar to a simplified belief network. Each node contributes a likelihood ratio. Risk factors multiply baseline odds.
Prior odds
Prior Odds = Prior Probability / (1 - Prior Probability)
Risk adjusted odds
Risk Odds = Prior Odds × Risk Multiplier
Positive evidence likelihood ratio
LR+ = Sensitivity / (1 - Specificity)
Negative evidence likelihood ratio
LR- = (1 - Sensitivity) / Specificity
Dependency correction
Adjusted Evidence LR = Raw Evidence LR ^ Dependency Factor
Final probability
Posterior Probability = Posterior Odds / (1 + Posterior Odds)
The dependency factor helps reduce double counting. Symptoms often share causes. Strong dependence lowers the combined evidence effect.
How to Use This Calculator
- Enter the starting TB probability as a percentage.
- Select exposure, immune, history, and setting risk factors.
- Choose the network dependence level.
- Set each symptom or test node as positive, negative, or unknown.
- Edit sensitivity and specificity values when better local estimates exist.
- Press the calculate button.
- Review the posterior probability and sensitivity band.
- Download CSV or PDF output for documentation.
Statistical Article About Hugin Network TB Probability Modeling
What The Model Does
A Hugin style network is a structured probability model. It uses nodes, states, and conditional evidence. This calculator follows that idea in a practical web form. The target node is TB probability. The evidence nodes include symptoms, imaging, exposure, and tests.
Why Odds Are Used
Bayesian updating is easier with odds. A prior probability becomes prior odds. Each new finding changes those odds with a likelihood ratio. A strong positive test raises the odds. A strong negative test lowers them. Unknown evidence has no effect.
Handling Dependent Evidence
Medical findings are rarely independent. Cough, fever, and weight loss may move together. Counting them as fully independent can overstate certainty. The dependence setting reduces the combined evidence effect. This gives a more cautious result.
Risk Factors And Prior Choice
The prior is the starting belief before evidence. It may reflect prevalence, screening group, or local data. Risk factors then adjust the odds. Close contact has a larger effect than low background exposure. Immune risk can also raise the starting estimate.
Interpreting Output
The result is not a diagnosis. It is a modeled probability based on entered assumptions. The sensitivity band shows how uncertainty may change the estimate. Review each likelihood ratio. A surprising result often comes from one strong input.
Good Statistical Practice
Use realistic inputs. Do not mix unrelated populations. Update test accuracy values when local validation is available. Keep unknown values as unknown. Avoid forcing weak evidence into the model. A transparent calculation is better than a hidden score.
Frequently Asked Questions
1. Is this calculator a medical diagnosis?
No. It is a statistical modeling tool. It supports structured probability review only. A clinician should interpret results with history, examination, testing, and local guidance.
2. What is the prior probability?
The prior probability is the starting estimate before evidence. It may come from prevalence, screening group data, known exposure level, or a professional baseline estimate.
3. Why does the calculator use likelihood ratios?
Likelihood ratios convert test accuracy into odds changes. A positive result uses LR+. A negative result uses LR-. This makes Bayesian updating easier and more transparent.
4. What does network dependence mean?
Network dependence means findings may overlap. Symptoms can share the same cause. The dependency factor reduces possible double counting when several related findings are entered.
5. Should unknown evidence be marked negative?
No. Unknown means not observed or not available. Negative means the finding was checked and absent. These states have different statistical meanings.
6. Can I change sensitivity and specificity values?
Yes. The defaults are editable. Use better values when you have local validation, published estimates, or lab-specific performance data.
7. Why does one test change the result strongly?
A test with high specificity can produce a large positive likelihood ratio. A test with high sensitivity can produce a useful negative likelihood ratio.
8. What exports are available?
The page provides CSV and PDF downloads. They include input assumptions, likelihood ratios, risk adjustment, posterior probability, and category output.