From a Per-Cell Hazard to a Cell-Loss Curve
A cell death rate describes the instantaneous risk of death for a cell that is still alive. It is not, by itself, the percentage of a population that will die over a day, a week, or any other finite interval. The distinction matters when translating a kinetic parameter into cell counts, comparing experiments with different observation times, or interpreting a reported fraction of cells lost.
Here we consider a defined cohort of cells and model death as its only way of leaving the living cohort. There is no cell division, influx, or transfer to another counted compartment during the interval. The death hazard is the same for all living cells and remains constant over time. Under these assumptions, the cohort follows first-order loss: the number dying per unit time is proportional to the number still alive.
Let \(N(t)\) be the number of living cells in the cohort at time \(t\), with \(N(0)=N_0\). The first-order loss model is
The minus sign indicates loss, and the factor \(N\) means that the absolute number lost per unit time decreases as fewer cells remain. The model treats \(N(t)\) as a deterministic population state or, for a finite stochastic cohort, as the expected number alive. The survival fraction is the fraction of the initial cohort still alive; the loss fraction is its complement.
Deriving Survival and Loss
Separate variables in the first-order loss equation and integrate from time 0 to time \(t\). This gives the survival curve directly.
For \(N(t)>0\), divide by \(N\) and multiply by \(dt\): \(\frac{dN}{N}=-k_d\,dt\).
With constant \(k_d\), integration from \(N_0\) to \(N(t)\) and from 0 to \(t\) gives \(\ln\!\left(\frac{N(t)}{N_0}\right)=-k_dt\).
The resulting survival and loss fractions are \(S(t)=e^{-k_dt}\) and \(F_{\mathrm{loss}}(t)=1-e^{-k_dt}\).
The integrated form also applies to any interval starting at time \(t_0\): conditional on being alive at \(t_0\), the fraction expected to survive an additional \(\Delta t\) is \(e^{-k_d\Delta t}\). The fraction of those cells that die during the interval is \(1-e^{-k_d\Delta t}\). Thus, the fraction lost depends on both the hazard and the interval length.
Two useful time summaries follow from the same survival curve. The half-life is the time at which \(S(t)=0.5\), so \(t_{1/2}=\ln(2)/k_d\). The mean lifetime is \(1/k_d\) for the exponential lifetime distribution implied by a constant hazard. Both have units of time; neither is a fraction lost per unit time.
Worked Example: Predicting Survival From a Death Hazard
Suppose a hypothetical cohort begins with \(N_0=10{,}000\) cells and has a constant death hazard \(k_d=0.18\ \mathrm{day}^{-1}\). Assume there is no division, influx, or other removal from the cohort. Find the expected number and fraction lost after 4 days.
The model gives \(N(4)=N_0e^{-k_d(4\ \mathrm{days})}\). The exponent is dimensionless because the units cancel:
The survival fraction is \(S(4)=e^{-0.72}\approx0.4868\). Therefore, the loss fraction is \(1-0.4868=0.5132\), or about 51.32%. The expected number lost is \(10{,}000-4{,}868\approx5{,}132\) cells. These are model-based expected quantities, not a claim that a finite experiment must yield exactly those counts.
Rate Versus Fraction Lost
The units and time scale distinguish a hazard from a finite-interval fraction. A hazard such as \(0.18\ \mathrm{day}^{-1}\) is an instantaneous conditional rate. A loss fraction such as 51.32% over 4 days is dimensionless and refers to a specified interval. Under a constant hazard, the hazard is constant but the absolute number dying per day declines as the cohort shrinks.
For a short interval, \(1-e^{-k_d\Delta t}\) is approximately \(k_d\Delta t\), because \(e^{-x}\approx1-x\) when \(x\) is small. This approximation is useful only when the product \(k_d\Delta t\) is small. For longer intervals, treating the hazard multiplied by time as the exact fraction lost can produce a substantial error.
Worked Example: Estimating the Hazard From Observed Cohort Survival
In a hypothetical experiment, a labeled cohort starts with \(N_0=120{,}000\) cells. After 6 days, \(42{,}000\) of the original cohort remain alive. Assume the measured cells represent the same cohort, with no division, influx, or other loss, and suppose a constant hazard is an appropriate model. Estimate \(k_d\).
The observed survival fraction is \(S(6)=42{,}000/120{,}000=0.35\). Since \(S(t)=e^{-k_dt}\), take natural logarithms and solve for the hazard:
The observed loss fraction over the 6 days is \(1-0.35=0.65\), or 65%. That is not the value of \(k_d\): the hazard is approximately \(0.1750\ \mathrm{day}^{-1}\), whereas 0.65 is a dimensionless fraction lost over 6 days. A single endpoint determines a constant hazard only under the stated model assumptions; it does not establish that the hazard was biologically constant.
Worked Example: Why a Hazard Is Not “Percent Lost Each Day”
Consider a hypothetical cohort with \(k_d=0.10\ \mathrm{day}^{-1}\). Calculate the fraction lost over 1 day and over 10 days, and compare those fractions with the hazard.
For 1 day, the exact fraction lost is
or about 9.516%. For 10 days, the calculation is
or about 63.21%. The hazard is \(0.10\ \mathrm{day}^{-1}\); it does not mean that exactly 10% of the original cohort dies each day. The 1-day approximation \(k_d\Delta t=0.10\) is close to, but not exactly, the 1-day loss fraction. Applying that approximation across 10 days would give 100%, which is not the model’s prediction.
Assumptions, Diagnostics, and Alternatives
The exponential survival curve is appropriate when a cell’s instantaneous death risk is proportional to the number of cells still at risk and is constant for each surviving cell over the interval. Applying it to an observed count trajectory additionally requires that the counted population represent the same cohort and that death be the only process changing its size. These conditions are model assumptions to assess, not conclusions guaranteed by an exponential-looking curve.
A useful diagnostic follows from taking logarithms: under the model, \(\ln N(t)=\ln N_0-k_dt\). Thus, log cell count plotted against time should follow a straight line with slope \(-k_d\), provided counts are positive and the assumptions hold. Systematic curvature, an apparent plateau, or different slopes in different portions of the trajectory signals that a single constant-hazard model may be inadequate. Measurement error, unrecorded influx or division, and changes in the cohort being counted can also produce departures. A straight line is compatible with the model but does not by itself prove its biological assumptions.
A key alternative is constant absolute loss: a fixed number of cells is removed per unit time, rather than a fixed per-cell hazard. That alternative produces a linear decline in cell count until the population reaches zero, unlike the exponential curve, whose absolute loss rate decreases with \(N(t)\). If a measured compartment receives cells or produces daughters while cells die, its total count is not simply a survival curve for the starting cohort. In that situation, a cohort label or an explicit accounting of the other flows is needed before a death hazard can be inferred from counts.
The model’s main strength is interpretability: one parameter determines the survival curve, and its units make comparisons across time scales possible. Its limitation is equally important: a constant hazard compresses potentially complex biology into a single rate. If the cohort is heterogeneous or conditions change, a fitted constant value may summarize the observed interval without representing a uniform per-cell mechanism.
Check Your Understanding
For each question, identify the relevant model quantity and show how the result follows from it.
- A cohort has \(k_d=0.25\ \mathrm{hour}^{-1}\). What fraction is expected to survive 2 hours, and what fraction is expected to be lost?
- A hypothetical labeled cohort has 80% survival after 3 days. Under a constant-hazard model, estimate \(k_d\) in \(\mathrm{day}^{-1}\).
- Explain why a hazard of \(0.05\ \mathrm{day}^{-1}\) does not imply that exactly 5% of the initial cohort is lost in every 1-day interval.
- A total cell count remains constant over time. Does this establish that the death hazard is zero? Explain what else could account for the observation.
Solutions. (1) \(S(2)=e^{-(0.25)(2)}=e^{-0.5}\approx0.6065\), so the expected survival fraction is 60.65% and the loss fraction is \(1-0.6065=0.3935\), or 39.35%. (2) \(k_d=-\ln(0.80)/(3\ \mathrm{days})=0.2231/3\approx0.07438\ \mathrm{day}^{-1}\). (3) The exact 1-day fraction lost is \(1-e^{-0.05}\approx0.04877\), or 4.877%; the hazard is a conditional instantaneous rate, not an exact percentage of the initial cohort. (4) No. Death may be balanced by influx or cell production, or the counted population may include changing flows. A constant total count alone does not identify the death hazard.