MPH Health Economics & Policy

Faculty of Health Sciences · CDU-HEA-MPH-900 · MSc · v1.0.0

The quantitative tools of health policy: the incremental cost-effectiveness ratio and net monetary benefit against a willingness-to-pay threshold, QALYs discounted at the reference-case rate, risk ratio, odds ratio and number needed to treat from a 2×2 table, and the predictive values and likelihood ratios of a diagnostic test.

Lineage

Conquest Designer University, 2026 — authored from the brief, the CEPH criteria and the NICE reference case; no prior institution is claimed

Final research project

Value for a health system: cost per QALY, discounted benefit, association and the worth of a test

Syllabus

Economic evaluation: incremental cost, incremental QALYs, ICER, dominance, net monetary benefit (NICE PMG36)
Outcomes: QALYs from utility weights, discounting at 3.5%
Epidemiology: risks, risk ratio with its confidence interval, risk difference, odds ratio, NNT
Screening and diagnosis: sensitivity, specificity, prevalence, predictive values, likelihood ratios

Reference

ICER: (C_new − C_comparator) ÷ (E_new − E_comparator)
Net monetary benefit: λ·ΔE − ΔC; λ = £20,000–£30,000 per QALY (NICE PMG36)
Discounting (NICE): 3.5% a year for costs and health effects; year 1 undiscounted here
Risk ratio 95% CI: exp(ln RR ± 1.96·√(1/a − 1/(a+b) + 1/c − 1/(c+d)))
PPV: sens·prev ÷ (sens·prev + (1 − spec)·(1 − prev))

Units

ICER and net monetary benefit · Discounted QALYs · Measures of association from a 2×2 table · Predictive values and likelihood ratios

Role when narrated: You are a public-health economist advising a health department. Say what the evidence shows and how certain it is, show the calculation, and separate the technical answer from the value judgement the decision-maker owns.

ICER and net monetary benefit · icer_nmb

Try with the example

What does each extra QALY from the new intervention cost, and is it worth it at the threshold?

Reference: NICE (2022) health technology evaluations manual PMG36: ICER against £20,000–£30,000 per QALY; net monetary benefit λΔE − ΔC

Returns

incrementalCost: ΔC
incrementalQaly: ΔE
icer: ΔC ÷ ΔE (null when ΔE = 0)
netMonetaryBenefit: λ·ΔE − ΔC
verdict: dominant, dominated, cost-effective or not cost-effective

The example

{
  "costNew": 24000,
  "costComparator": 15000,
  "qalyNew": 6.2,
  "qalyComparator": 5.8,
  "threshold": 20000
}

Agents: POST /faculty/mph-health-economics--policy/icer_nmb with JSON {pan, …inputs} and the x-402-payment header. Manifest.

Discounted QALYs · discounted_qalys

Try with the example

How many QALYs does this stream of health-state utilities yield, undiscounted and discounted at the reference-case rate?

Reference: NICE PMG36 reference case: health effects discounted at 3.5% a year; QALY = Σ utility × time

Returns

lifeYears: Years counted
undiscountedQalys: Σ utilities
discountedQalys: Σ u_t ÷ (1 + r)^t, t = 0 for year 1

The example

{
  "utilities": [
    0.85,
    0.82,
    0.8,
    0.78,
    0.75
  ],
  "discountRate": 0.035
}

Agents: POST /faculty/mph-health-economics--policy/discounted_qalys with JSON {pan, …inputs} and the x-402-payment header. Manifest.

Measures of association from a 2×2 table · two_by_two

Example filled in — consult below

In this cohort, how much more (or less) likely is the outcome among the exposed, and how many must be exposed for one extra (or one fewer) case?

Reference: Rothman, Greenland & Lash, Modern Epidemiology: risk ratio (log-scale 95% CI), risk difference, odds ratio; NNT = 1/|RD|

Returns

riskExposed: a ÷ (a + b)
riskUnexposed: c ÷ (c + d)
riskRatio: RR
riskRatioLower95: Lower 95% bound of RR
riskRatioUpper95: Upper 95% bound of RR
riskDifference: Risk exposed − risk unexposed
oddsRatio: (a·d) ÷ (b·c)
numberNeeded: 1 ÷ |RD| (to harm if RD > 0, to treat if RD < 0)

The example

{
  "exposedCases": 30,
  "exposedNonCases": 70,
  "unexposedCases": 15,
  "unexposedNonCases": 85
}

Agents: POST /faculty/mph-health-economics--policy/two_by_two with JSON {pan, …inputs} and the x-402-payment header. Manifest.

Predictive values and likelihood ratios · diagnostic_accuracy

Try with the example

In a population with this prevalence, how likely is a positive (or negative) result to be right?

Reference: Bayes’ theorem applied to a diagnostic test; likelihood ratios LR+ = sens/(1 − spec), LR− = (1 − sens)/spec

Returns

ppv: P(disease | +)
npv: P(no disease | −)
positiveLikelihoodRatio: LR+
negativeLikelihoodRatio: LR−

The example

{
  "sensitivity": 0.9,
  "specificity": 0.95,
  "prevalence": 0.02
}

Agents: POST /faculty/mph-health-economics--policy/diagnostic_accuracy with JSON {pan, …inputs} and the x-402-payment header. Manifest.