Half-Life

Half-Life: Concept Architecture

Ask what quantity is halving

A half-life is the elapsed time required for a specified quantity to fall to half its starting value under a stated process. The term appears in drug concentrations, biological materials, and assumptions about how a treatment effect wanes. This page derives the simple exponential case, works through concentration and health-economic examples, and explains when one half-life cannot be transferred to another outcome.

Derive the constant-fraction decay rule

Let $Q(t)$ be a nonnegative quantity measured from a defined starting point. Under first-order exponential decay with a constant proportional decay rate $k>0$, the quantity follows $Q(t)=Q_0e^{-kt}$. Setting it equal to $Q_0/2$ and solving gives:

$$t_{1/2}=\frac{\ln 2}{k}, \qquad Q(t)=Q_0\,2^{-t/t_{1/2}}.$$

Here $k$ has inverse-time units and $t_{1/2}$ has time units. Each additional half-life halves the amount remaining: after two half-lives one quarter remains, and after three one eighth. The rule depends on proportional rather than constant-amount loss and on a stable rate over the interval. A single observed decline does not establish those assumptions or a unique half-life for all future times.

Work through an illustrative drug concentration

Suppose a measured plasma concentration is 80 mg/L at the beginning of a terminal elimination phase and that phase has an eight-hour half-life. The exponential model predicts 40 mg/L after eight hours and 20 mg/L after 16 hours, assuming no further dose and the same terminal-phase conditions. The arithmetic shows concentration halving, not the fraction of patients who have recovered or the fraction of treatment benefit remaining.

In a simple one-compartment linear model after distribution, an elimination rate constant can be related to clearance and volume of distribution as $k=CL/V$, giving $t_{1/2}=\ln(2)V/CL$. Clearance has volume-per-time units and volume of distribution has volume units, so the ratio has the inverse-time units required for $k$. Multicompartment drugs may have an early distribution phase and a different terminal slope, and a terminal half-life does not alone determine time to steady state or clinical effect. Pharmacokinetic research on terminal slopes examines the conditions behind a single exponential phase, while analysis of distribution volumes shows why simple steady-state rules may fail.

Separate exposure from response

Drug concentration, target occupancy, biomarker response, symptom relief, and durable clinical benefit can decline on different timescales. An immune response or irreversible biological action may persist after plasma drug is low; conversely, concentration above a chosen level does not guarantee benefit. Define the quantity and its measurement method before calling any interval a treatment's half-life.

Nonlinear elimination, time-varying clearance, repeated dosing, active metabolites, or a multiphasic curve can make a single constant half-life unsuitable. For repeated doses, concentration reflects both new input and prior doses, so it need not halve between clock times. State whether the estimate concerns an observed terminal concentration slope, a model parameter, or an independently measured clinical-effect duration.

Model waning treatment benefit explicitly

Health economic models sometimes assume that an effect seen during trial follow-up becomes smaller afterward. If the treatment effect is represented by a log hazard ratio, an illustrative exponential waning assumption is:

$$\log HR(t)=\log HR_0\,2^{-t/h}, \qquad HR(t)=\exp\!\left[\log HR_0\,2^{-t/h}\right],$$

where $h$ is the assumed half-life of the log hazard-ratio effect measured from the start of waning. If $HR_0=0.60$ and $h=2$ years, then at two years $HR(2)=\exp[\log(0.60)/2]=\sqrt{0.60}\approx0.775$, moving toward no relative effect at $HR=1$. Multiplying the hazard ratio itself by one half would give 0.30, moving in the wrong direction for a protective effect. This is a demonstration of one specified waning scale, not a rule that clinical treatment effects must follow it.

The time origin, quantity that wanes, comparator hazard, and whether waning starts at treatment stop or another point must all be stated. An absolute risk difference or excess hazard requires a different equation and can produce different projected survival and quality-adjusted life years. Methods research on treatment-effect waning highlights that marginal and conditional effect measures can behave differently under extrapolation; an HTA review documents the importance of supporting assumptions.

Distinguish half-life from median survival

For an exponential time-to-event model with a constant event hazard $\lambda$, the event-free survival probability is $S(t)=e^{-\lambda t}$, so its median event time is $\ln(2)/\lambda$. The matching algebra does not make a patient's remaining lifetime a decaying substance. It says that half the defined population would have had the specified event by that time under that survival model and its censoring assumptions.

If the hazard changes with time or competing events alter the estimand, there may be no single constant-rate half-life. A Kaplan–Meier median is read from an estimated survival curve when it reaches 0.5 and does not require imposing an exponential model. Keep medication elimination, treatment-effect waning, and population event-time summaries separate even when each uses the word “half.”

Check what the decision depends on

Report the quantity, units, time origin, model family, estimated or assumed rate, data range, uncertainty, and the clinical or economic outcome affected. Test whether plausible faster, slower, or no-waning assumptions change projected costs, survival, quality-adjusted life years, or the decision. Validate against follow-up and relevant external evidence rather than treating a convenient half-life as measured fact.

A half-life describes a specific dynamic under a specific model; it is not a universal duration of action. When evidence cannot support a stable halving rule, use a more appropriate time-varying model or present the half-life as an explicit scenario rather than silently embedding it in the base case.