Early Health Technology Assessment, Headroom Pricing and Benefit-Package Decisions

อ่านฉบับภาษาไทย (Thai version)

Abstract

A company has built a sensor that warns a heart-failure nurse days before fluid fills a patient's lungs. Its trial has not been run, yet the company must decide now whether to fund one, and at what price. Early health technology assessment answers with stated guesses instead of trial results. We value the guessed gain at the health scheme's threshold, the most it will pay per year of full health, 160,000 THB: 0.4 of such a year is worth 64,000 THB. Adding the 10,000 THB the sensor saves gives 74,000 THB, the highest good-value price: its price ceiling. The company hoped for 100,000 THB, so the price needs a 26% cut, or the trial must show at least 0.56 of such a year. If the care the money pushes out buys a year of full health for 100,000 THB, even a price of 60,000 THB loses health. The ceiling plans the trial and the price talks; it does not promise coverage.

🔍 Click to enlarge Click anywhere to close

A sensor that is not finished yet

A small medical-device company in Chiang Mai has built a sensor for people with heart failure. A nurse places it under the skin of the chest during a short clinic visit. Every morning it sends the heart-failure nurse a reading that rises days before fluid builds up in the lungs. The nurse can then change the medicines by phone, before the patient needs a hospital bed.

In a first study of 30 patients the sensor worked and patients kept it in, but nobody yet knows how many hospital stays it prevents. A trial large enough to show that would cost the company 44 million THB (Thai baht) and take three years. The company hopes to sell each sensor to the national health scheme, the payer in this story, for 100,000 THB.

Before it spends 44 million THB, the company asks two questions. If the trial shows what the team expects, would the scheme pay 100,000 THB per sensor? If not, what price, and what trial result, would it need?

A common reading of an early model goes one step too far: if the model finds a price that is good value, the scheme should pay that price. It should not. An early model gives a ceiling that holds only while its guesses hold, and a list of what the trial must show. The decision to pay comes later and weighs more than value.

Who, compared with what, at which step of care

Every number below depends on three choices made first. The patients are adults sent home after a hospital stay for heart failure. The comparator, what they would get without the sensor, is usual care: a clinic visit every three months and phone calls from the nurse. The sensor is added to usual care after discharge, and it replaces nothing.

Change any of the three and the answer changes. If the comparator were a daily weighing programme that already prevents many hospital stays, the sensor would add less over it, and every price below would fall.

Working out what a technology still in development must achieve, and at what price it could be good value, is called early health technology assessment, or early HTA [1]. Because the trial has not been run, it works from scenarios: stated guesses for each missing number, each with its source and a range to test.

From the first study and similar devices, the team's main guess is that the sensor adds, over each patient's remaining life, as much health as 0.4 of a year in full health. Health is counted in QALYs, quality-adjusted life years: a year of life weighted by its quality, from 1 for full health to 0 for death. Over 10 years, a state valued at 0.70 counts as 7.0 QALYs, so restoring full health gains 3.0; the full lesson is in QALY, DALY and the Utilities Behind Them: What Is Actually Being Measured.

So the 0.4 QALY comes from fewer hospital stays and better days at home.

The most the scheme could pay for the sensor

The scheme pays for a new treatment only if the health it adds is worth the extra money. Suppose the most it is willing to pay for each QALY a treatment adds is 160,000 THB; that benchmark is called its threshold. We use 160,000 THB as an illustration, not as a confirmed current national figure.

The company needs the price at which the sensor sits exactly on that threshold. Below that price, each QALY the sensor adds costs the scheme less than 160,000 THB; above it, more. Two kinds of money set that price: what the health is worth, and what else changes in the scheme's spending.

Fitting the sensor and reading it each day add 8,000 THB per patient of nurse and clinic time. Fewer hospital stays save 18,000 THB per patient over the rest of life. Write every change in the scheme's spending with its sign: +8,000 THB for the fitting and reading, −18,000 THB for the stays avoided. A minus sign means money the scheme no longer spends.

The name: headroom

This highest price is called the sensor's headroom, or its value-based price ceiling: the most a payer, here the scheme, could pay and still get value at its threshold [2]. Write λ for the threshold per QALY, ΔE for the health gain per patient and ΔC_other for the signed change in all other costs per patient. Let q be the number of sensors each patient needs. At a price P per sensor the extra cost per patient is q × P + ΔC_other, so the ceiling P_max per sensor is:

$$P_{max}=\frac{\lambda\times\Delta E-\Delta C_{other}}{q}$$

Each patient needs one sensor, so q = 1 and the ceiling is 0.4 × 160,000 − (−10,000) = 74,000 THB. Subtracting a negative number adds it, which is why a saving raises the ceiling. The price is for the sensor alone; usual care contains no sensor to subtract. If usual care used a device, the device no longer needed would enter ΔC_other as a saving.

The sign is where early models most often slip. If the 10,000 THB saving is written as a plus and then subtracted like a cost, the ceiling becomes 64,000 − 10,000 = 54,000 THB, which is 20,000 THB too low. One rule prevents it: every change in spending carries its own sign, and the formula always subtracts it.

f <- function(x, d = 0) formatC(x, format = "f", digits = d, big.mark = ",")
threshold <- 160000
qaly_gain <- 0.400
fitting <- 8000
stays <- -18000
value <- threshold * qaly_gain
other <- fitting + stays
ceiling <- value - other
cat("10 years at 0.70 =", f(10 * 0.70, 1), "QALYs; full health adds", f(10 * 0.30, 1), "\n")
cat("Health gain valued at the threshold:", f(value), "THB; other cost change:", f(other), "THB\n")
cat("Price ceiling:", f(ceiling), "THB; per QALY there:", f((ceiling + other) / qaly_gain), "THB\n")
cat("Saving taken as a cost:", f(value - abs(other)), "THB, too low by", f(2 * abs(other)), "THB\n")

10 years at 0.70 = 7.0 QALYs; full health adds 3.0
Health gain valued at the threshold: 64,000 THB; other cost change: -10,000 THB
Price ceiling: 74,000 THB; per QALY there: 160,000 THB
Saving taken as a cost: 54,000 THB, too low by 20,000 THB

The asking price, the margin and the trial

The company hoped for 100,000 THB, which is 26,000 THB above the ceiling. To reach the ceiling the price would need a cut of 26,000 ÷ 100,000 = 26%. At 100,000 THB the scheme would pay 100,000 − 10,000 = 90,000 THB more per patient for 0.4 QALY, or 225,000 THB per QALY, well over its threshold. Put another way, the health is worth 64,000 THB and the extra cost is 90,000 THB, so health value minus extra cost is 64,000 − 90,000 = −26,000 THB: below zero by exactly the 26,000 THB that the price is over the ceiling.

The ceiling is a price, not a profit. Making one sensor costs the company 30,000 THB, so at the ceiling it keeps 74,000 − 30,000 = 44,000 THB per sensor. From that margin it must repay the 44 million THB trial, which takes 1,000 sensors. Its other business costs come on top.

So the company's question is a go or no-go: is a price of 74,000 THB or less still worth the trial? That is a business decision, made now, by the company. The scheme's decision to pay comes later, after the trial, and asks more than value.

f <- function(x, d = 0) formatC(x, format = "f", digits = d, big.mark = ",")
qaly_gain <- 0.400
other <- 8000 - 18000
ceiling <- 160000 * qaly_gain - other
offer <- 100000
extra <- offer + other
cut <- 100 * (1 - ceiling / offer)
cat("Offer over the ceiling:", f(offer - ceiling), "THB, a price cut of", f(cut, 2), "%\n")
cat("Extra cost at the offer:", f(extra), "THB, or", f(extra / qaly_gain), "THB per QALY\n")
cat("Health value minus extra cost at the offer:", f(160000 * qaly_gain - extra), "THB\n")
cat("Margin per sensor over a making cost of 30,000 THB:", f(ceiling - 30000), "THB\n")
cat("Sensors that pay back a 44 million THB trial:", f(44000000 / (ceiling - 30000)), "\n")

Offer over the ceiling: 26,000 THB, a price cut of 26.00 %
Extra cost at the offer: 90,000 THB, or 225,000 THB per QALY
Health value minus extra cost at the offer: -26,000 THB
Margin per sensor over a making cost of 30,000 THB: 44,000 THB
Sensors that pay back a 44 million THB trial: 1,000

What the trial must show

The 0.4 QALY is a guess, so an early model tries other values too. With the same savings, a gain of 0.2 QALY gives a ceiling of 42,000 THB and a gain of 0.6 QALY gives 106,000 THB. Each extra 0.1 QALY adds 0.1 × 160,000 = 16,000 THB to the ceiling. At 0.2 QALY the company's margin falls to 42,000 − 30,000 = 12,000 THB per sensor, so repaying the 44 million THB trial would take 44 million ÷ 12,000 = about 3,667 sensors.

The savings matter as well. If the sensor prevents no hospital stays, only the +8,000 THB of fitting and reading is left. At 0.4 QALY the ceiling then falls to 64,000 − 8,000 = 56,000 THB.

So the 100,000 THB the company hoped for is good value only if the trial shows, with the stays avoided, a gain of at least (100,000 − 10,000) ÷ 160,000 = about 0.56 QALY. The trial must therefore measure the two numbers that move the ceiling most: the health gain and the hospital stays avoided. A smaller study that shows only how well the sensor detects fluid would not answer the scheme's question.

f <- function(x, d = 0) formatC(x, format = "f", digits = d, big.mark = ",")
threshold <- 160000
gains <- c(0.200, 0.400, 0.600)
other <- 8000 - 18000
ceilings <- threshold * gains - other
margin_low <- ceilings[1] - 30000
and <- function(x) paste(x, collapse = " and ")
cat("Ceilings for gains of", and(f(gains, 1)), "QALY:", and(f(ceilings)), "THB\n")
cat("Each extra 0.1 QALY adds", f(threshold * 0.1), "THB to the ceiling\n")
cat("At 0.2 QALY: margin", f(margin_low), "THB; sensors to repay", f(44000000 / margin_low), "\n")
cat("Gain the asking price of 100,000 THB needs:", f((100000 + other) / threshold, 2), "QALY\n")
cat("Ceiling at 0.4 QALY if no hospital stays are avoided:", f(threshold * 0.400 - 8000), "THB\n")

Ceilings for gains of 0.2 and 0.4 and 0.6 QALY: 42,000 and 74,000 and 106,000 THB
Each extra 0.1 QALY adds 16,000 THB to the ceiling
At 0.2 QALY: margin 12,000 THB; sensors to repay 3,667
Gain the asking price of 100,000 THB needs: 0.56 QALY
Ceiling at 0.4 QALY if no hospital stays are avoided: 56,000 THB

A price under the ceiling can still cost health

The threshold of 160,000 THB is a choice: the most the scheme says it is willing to pay for a QALY. What the scheme's money buys elsewhere is a different number. Money for the sensor comes out of other care the scheme would have paid for, such as clinic visits or other medicines, and that care also produces health. Call it the care pushed out.

Suppose the care pushed out buys a QALY for every 100,000 THB, an illustrative figure. Suppose too that the scheme talks the price down to 60,000 THB, well under the ceiling. Each patient then costs the scheme 60,000 − 10,000 = 50,000 THB extra, money that would have bought 50,000 ÷ 100,000 = 0.5 QALY of other care. The sensor adds 0.4 QALY, so health changes by 0.4 − 0.5 = −0.1 QALY per patient fitted: a net loss.

The health lost in the care pushed out is called the health opportunity cost. Health gained minus health pushed out is the net health benefit [3]; both come with a province's numbers in the article linked next: Cost-Effective Is Not Affordable, and Efficient Is Not Fair: Budget Impact and Equity.

With k the cost per QALY of the care pushed out, net health is zero at a price of k × ΔE − ΔC_other = 100,000 × 0.4 − (−10,000) = 50,000 THB.

So the sensor has two ceilings: 74,000 THB at the threshold the scheme states, and 50,000 THB at what its money buys elsewhere. The threshold is the scheme's choice and k is a fact of its health system, so neither can stand in for the other. Nobody knows k exactly, and the gap between the two ceilings is a judgment the scheme must make in the open, not a detail of the model.

f <- function(x, d = 0) formatC(x, format = "f", digits = d, big.mark = ",")
k <- 100000
qaly_gain <- 0.400
other <- 8000 - 18000
price <- 60000
pushed_out <- (price + other) / k
cat("Extra cost at a price of", f(price), "THB:", f(price + other), "THB\n")
cat("Health pushed out:", f(pushed_out, 1), "QALY; net:", f(qaly_gain - pushed_out, 1), "QALY\n")
cat("Price where net health is zero:", f(k * qaly_gain - other), "THB\n")

Extra cost at a price of 60,000 THB: 50,000 THB
Health pushed out: 0.5 QALY; net: -0.1 QALY
Price where net health is zero: 50,000 THB

From a ceiling to a benefit-package decision

Read first on this site:

• How economic evaluation feeds national drug selection in Thailand
• How a fund counts the people who qualify, year by year, before it counts money

A benefit package is the list of treatments a health scheme pays for. A decision to add the sensor weighs value beside other criteria, and stating each criterion and its weight in the open is the idea behind multiple criteria decision analysis [4].

The ceiling is per patient, but the bill is per year. If 2,000 patients a year receive the sensor at 60,000 THB, the scheme pays 2,000 × 60,000 = 120 million THB a year for sensors alone. It pays at the time of fitting, while the hospital stays avoided come over the following years. Whether the scheme can find that money each year is the budget question, answered for a province in the article linked next: Cost-Effective Is Not Affordable, and Efficient Is Not Fair: Budget Impact and Equity.

Uptake, the number of patients who get the sensor, moves this bill but not the ceiling. With 1,000 or 3,000 patients a year the sensors cost 60 or 180 million THB, while the ceiling per patient stays 74,000 THB.

At 60,000 THB, if the care pushed out buys a QALY for 100,000 THB, each patient loses 0.1 QALY, so 2,000 patients cost the population 2,000 × 0.1 = 200 QALYs a year.

f <- function(x, d = 0) formatC(x, format = "f", digits = d, big.mark = ",")
patients <- 2000
price <- 60000
net_per_patient <- 0.400 - (price + 8000 - 18000) / 100000
cat("Sensors for", f(patients), "patients a year:", f(patients * price / 1e6), "million THB\n")
uptake <- c(1000, 3000)
and <- function(x) paste(x, collapse = " and ")
cost <- and(f(uptake * price / 1e6))
cat("Sensors for", and(f(uptake)), "patients a year:", cost, "million THB\n")
cat("Net health, care pushed out at 100,000 per QALY:", f(patients * net_per_patient), "QALYs\n")

Sensors for 2,000 patients a year: 120 million THB
Sensors for 1,000 and 3,000 patients a year: 60 and 180 million THB
Net health, care pushed out at 100,000 per QALY: -200 QALYs

Two more checks, then conditional options

The ceiling answers the value check and the yearly bill answers the cost check; two more checks come before a yes. First, can the system deliver it, with enough heart-failure nurses to read 2,000 new patients' readings every morning? Second, is it fair? The sensor needs a mobile signal and a nearby heart-failure clinic, so rural patients, who live furthest from a hospital bed, may get it last [5].

An early model cannot make these choices, but it can turn them into conditional options. Negotiate: pay only at or under the ceiling of 74,000 THB, or nearer 50,000 THB if the scheme accepts the lower figure for k.

Target: pay first for patients with the most to gain, such as those with two hospital stays last year. If the trial shows 0.6 QALY for them and the same 10,000 THB saving, their ceiling is 0.6 × 160,000 + 10,000 = 106,000 THB.

Pay while collecting evidence, or wait: cover the sensor in hospitals that record every hospital stay and review in two years, or decide only after the trial.

Each option needs monitoring criteria and a monitoring trigger: a result, agreed in advance, that reopens the decision at the review. Count the hospital stays avoided per patient and value them in THB; the model assumed they save 18,000 THB per patient. Also count the days with a reading sent, spending against the forecast, and who receives the sensor, town and rural. If the stays avoided fall well short, the price falls or the cover stops at the review.

The series, in learning order

1. Health Economics Series Guide: Read the Evidence Before the Costs
2. Incidence, Prevalence and the Right Denominator: The Epidemiology Beneath Cost-of-Illness and Markov Models
3. Design, Model and Estimand: What Each Study Design Can Actually Estimate
4. Risk Ratio, Odds Ratio, Risk Difference and NNT: One 2×2 Table, Four Answers
5. The Hazard Ratio Is Not a Risk Ratio: Proportional Hazards, Restricted Mean Survival Time and Model Inputs
6. Chance, Bias and Confounding: Why an Association Is Not Yet a Treatment Effect
7. Beyond the Evidence Pyramid: GRADE Certainty and Choosing Effectiveness Inputs for Economic Models
8. From Sensitivity to Cost per Case Detected: Predictive Values, Prevalence and the Economics of Screening
9. Whose Cost, in Which Year? Perspective, Opportunity Cost, Inflation and Discounting
10. QALY, DALY and the Utilities Behind Them: What Is Actually Being Measured
11. ICER, Dominance and the Cost-Effectiveness Plane: Reading the Thai Threshold of 160,000 Baht per QALY
12. Decision Trees, Markov Cohorts and Probabilistic Sensitivity Analysis: How a Model Turns Evidence into a Decision
13. Cost-Effective Is Not Affordable, and Efficient Is Not Fair: Budget Impact and Equity
14. Parameter Sourcing, Reference Cost Lists and Utility Catalogues When Local Data Are Sparse
15. Early Health Technology Assessment, Headroom Pricing and Benefit-Package Decisions (this article)
16. Communicating Health Economic Evidence: From Model Outputs to Policy Briefs

Related in the wiki: ICER, Dominance and the Cost-Effectiveness Plane: Reading the Thai Threshold of 160,000 Baht per QALY · Decision Trees, Markov Cohorts and Probabilistic Sensitivity Analysis: How a Model Turns Evidence into a Decision · Cost-Effective Is Not Affordable, and Efficient Is Not Fair: Budget Impact and Equity