Why Averaging Parameters Is Not Averaging Elimination
In a population model, individuals can differ in both \(V_{\max}\) and \(K_m\). At a specified concentration, each individual therefore has their own elimination rate. The average of those rates is a population quantity, but it is not generally the rate obtained by inserting average parameter values into the Michaelis–Menten equation. This distinction matters whenever a model-based prediction is meant to describe the population’s average elimination rather than a representative individual.
The comparison is clearest when concentration is held fixed across individuals. If concentrations also vary, the population average depends on the joint distribution of concentration and individual parameters. The examples below first isolate parameter heterogeneity, then show why concentration heterogeneity adds another layer.
In a finite population or a discrete example, the expectation is an arithmetic average over individuals. In a continuous random-effects model, it is an integral over the joint distribution of \(V_{\max}\) and \(K_m\). If random effects are used to define individual parameters, the population average must be computed using the resulting parameter distribution, including any dependence between the parameters. A zero-mean random effect on a log scale does not generally make the corresponding parameter’s arithmetic mean equal to its typical value.
A Useful Result When Only \(K_m\) Varies
A simple case gives a precise direction for the difference. Suppose everyone has the same \(V_{\max}=V\), the same concentration \(C>0\), and positive, heterogeneous \(K_m\). The rate as a function of \(K_m\) is \(VC/(K_m+C)\). The reciprocal term is convex for positive \(K_m\), so averaging it gives at least its value at the average \(K_m\).
By Jensen’s inequality, a convex function evaluated at a random variable has an expectation at least as large as the function evaluated at that variable’s expectation. Multiplying this inequality by the positive constant \(VC\) gives:
Thus, with fixed \(V_{\max}\) and a common positive concentration, the average individual rate is at least the rate at average \(K_m\). The inequality is strict when \(K_m\) varies across individuals; equality holds when \(K_m\) is constant. This result does not establish a universal direction when \(V_{\max}\) also varies.
Worked Example: Fixed Vmax and Heterogeneous Km
Consider two equally represented hypothetical individuals at \(C=10\ \text{mg/L}\). Both have \(V_{\max}=100\ \text{mg/h}\), while their \(K_m\) values are \(10\ \text{mg/L}\) and \(90\ \text{mg/L}\). Compare their average rate with the rate at the average parameters.
Average the individual rates. For the first individual, \(R_1(10)=100(10)/(10+10)=1{,}000/20=50.0\ \text{mg/h}\). For the second, \(R_2(10)=100(10)/(90+10)=1{,}000/100=10.0\ \text{mg/h}\). The population-average rate is \((50.0+10.0)/2=30.0\ \text{mg/h}\).
Evaluate at the average parameters. The mean \(K_m\) is \((10+90)/2=50\ \text{mg/L}\), and mean \(V_{\max}\) remains \(100\ \text{mg/h}\). Therefore, \(R(10;100,50)=100(10)/(50+10)=1{,}000/60=16.67\ \text{mg/h}\).
The average individual rate, \(30.0\ \text{mg/h}\), exceeds the rate at the average parameters, \(16.67\ \text{mg/h}\), as the convexity result predicts. The difference is \(13.33\ \text{mg/h}\). The calculation compares rates at the same concentration; it does not claim that either rate describes every individual.
When Both Parameters Vary
When \(V_{\max}\) varies as well, the expectation involves a product: \(E[V_{\max}C/(K_m+C)]\). If \(V_{\max}\) and \(K_m\) are dependent, their joint distribution matters. Even with independence, the fixed-\(V_{\max}\) Jensen result cannot be applied without accounting for the variation in \(V_{\max}\).
For \(C>0\), write the expectation using a covariance decomposition:
The function \(1/(K_m+C)\) decreases as \(K_m\) increases. Consequently, dependence between \(V_{\max}\) and \(K_m\) affects the covariance term, and that term can change the comparison with the rate at the average parameters. A positive association between \(V_{\max}\) and \(K_m\), for instance, tends to pair larger values of \(V_{\max}\) with smaller values of the reciprocal term. The net direction still depends on the full joint distribution and the amount of heterogeneity.
Worked Example: Heterogeneity Can Reverse the Direction
Consider two equally represented hypothetical individuals at \(C=10\ \text{mg/L}\). Individual 1 has \(V_{\max}=20\ \text{mg/h}\) and \(K_m=10\ \text{mg/L}\); individual 2 has \(V_{\max}=180\ \text{mg/h}\) and \(K_m=90\ \text{mg/L}\). Calculate both population summaries.
The individual rates are \(R_1(10)=20(10)/(10+10)=200/20=10.0\ \text{mg/h}\) and \(R_2(10)=180(10)/(90+10)=1{,}800/100=18.0\ \text{mg/h}\). Their average is \((10.0+18.0)/2=14.0\ \text{mg/h}\).
The average parameters are \(E[V_{\max}]=(20+180)/2=100\ \text{mg/h}\) and \(E[K_m]=(10+90)/2=50\ \text{mg/L}\). The rate at those averages is \(100(10)/(50+10)=1{,}000/60=16.67\ \text{mg/h}\).
Here the population-average individual rate, \(14.0\ \text{mg/h}\), is lower than the rate at average parameters, \(16.67\ \text{mg/h}\). This does not contradict the earlier inequality: \(V_{\max}\) is not fixed. In this example the higher \(V_{\max}\) occurs with the higher \(K_m\), so it also occurs with a smaller value of \(1/(K_m+C)\). The comparison cannot be determined from the means alone.
Concentration Variation Changes the Average Too
So far, individuals were compared at a common concentration. In a population time course, individuals may have different concentrations as well as different parameters. The relevant average rate is then \(E[V_{\max,i}C_i/(K_{m,i}+C_i)]\), where the distribution of \(C_i\) may be related to the individual parameters. It is generally not equal to the rate evaluated at average concentration and average parameters.
Worked Example: Different Individual Concentrations
Suppose two equally represented hypothetical individuals share \(V_{\max}=100\ \text{mg/h}\) and \(K_m=20\ \text{mg/L}\), but their concentrations are \(10\ \text{mg/L}\) and \(30\ \text{mg/L}\). Compare the average rate with the rate at the average concentration.
At \(C_1=10\ \text{mg/L}\), \(R_1=100(10)/(20+10)=1{,}000/30=33.33\ \text{mg/h}\). At \(C_2=30\ \text{mg/L}\), \(R_2=100(30)/(20+30)=3{,}000/50=60.0\ \text{mg/h}\). The average individual rate is \((33.33+60.0)/2=46.67\ \text{mg/h}\).
The average concentration is \((10+30)/2=20\ \text{mg/L}\). Evaluating the rate there gives \(100(20)/(20+20)=2{,}000/40=50.0\ \text{mg/h}\). Thus the average individual rate is \(3.33\ \text{mg/h}\) lower. For fixed positive \(V_{\max}\) and \(K_m\), the rate is concave as a function of concentration, so averaging concentrations before evaluating the rate can exceed the average rate. This is a separate source of averaging differences from heterogeneity in \(K_m\).
Computing and Checking Population Averages
For a discrete population, calculate each individual’s rate at their own parameters and concentration, then average those rates using the intended population weights. For a continuous random-effects distribution, the corresponding target is an expectation over the joint distribution. Numerical integration or draws from that distribution can approximate it. The parameterization and dependence structure must match the population being described; using independent draws when the model specifies correlated random effects changes the target.
State whether the comparison is at a common concentration or at individual concentrations, and identify the population weights or random-effects distribution.
Apply \(R_i(C_i)=V_{\max,i}C_i/(K_{m,i}+C_i)\) before averaging. Keep the concentration and rate units consistent.
Calculate the relevant parameter and, if applicable, concentration averages, then evaluate the rate. Do not treat this result as the average rate.
Compare direct averaging with numerical integration or a sufficiently large simulation from the same joint distribution. Check that simulated values preserve parameter correlations and that the numerical average is stable as simulation size increases.
These checks assess whether the population average has been calculated as intended; they do not establish that the chosen population model is correct. Large differences between the two summaries are not automatically a model failure. They can be an expected consequence of strong heterogeneity and the nonlinear rate law. Conversely, a plug-in calculation presented as a population average is a reporting error even if the numerical difference happens to be small.
Interpretation, Alternatives, and Limitations
There are several legitimate summaries, but they answer different questions. The rate at average parameters describes the rate of a hypothetical individual with those parameter values at the specified concentration. The average individual rate describes the mean of the individual rates in the population. A typical individual, such as one defined by typical random effects, is another target and need not have the population arithmetic means of \(V_{\max}\) and \(K_m\). For time-course predictions, averaging individual predictions preserves parameter heterogeneity; solving a single model at average parameters does not generally reproduce that average trajectory.
| Calculation | What it summarizes | Main limitation |
|---|---|---|
| Rate at average parameters | A hypothetical individual assigned the parameter averages | Not generally the population-average rate |
| Average individual rate | The arithmetic mean rate over the specified population | Requires the relevant parameter and concentration distribution |
| Rate for typical random effects | A model-defined typical individual | May differ from the arithmetic population average |
| Average simulated or integrated predictions | A population summary that retains modeled heterogeneity | Depends on the model distribution and accurate numerical computation |
A common mistake is to calculate the average \(V_{\max}\) and average \(K_m\), insert them into the Michaelis–Menten equation, and label the result “mean population elimination.” The correction is to distinguish a plug-in rate from the average of individual rates and calculate the latter by averaging or integrating the individual rate law. Another mistake is to infer the direction of the difference solely from Jensen’s inequality. That direction follows in the fixed-\(V_{\max}\), common-concentration case, but not in general when both parameters vary or when concentrations vary.
Population averages also depend on the chosen population and the time or concentration at which they are evaluated. Averages from one covariate mix or parameter distribution need not describe a population with a different mix. Finally, the average rate at a time point is not by itself a complete description of the population’s concentration-time behavior: individual concentrations and parameters can co-evolve, so an average rate at one instant does not generally determine an average trajectory.
Check Your Understanding
Answer each question using the distinction between an average rate and a rate evaluated at average values.
- With common \(V_{\max}\) and common positive concentration, does heterogeneity in positive \(K_m\) make the average individual rate at least as large as, or at most as large as, the rate at average \(K_m\)? State the mathematical reason.
- Two equally weighted individuals have \(V_{\max}=50\ \text{mg/h}\) and \(K_m=10\ \text{mg/L}\) for the first, and \(V_{\max}=150\ \text{mg/h}\) and \(K_m=30\ \text{mg/L}\) for the second. At \(C=10\ \text{mg/L}\), calculate the average individual rate and the rate at average parameters.
- Why does the fixed-\(V_{\max}\) Jensen result not establish the direction of the difference when both \(V_{\max}\) and \(K_m\) vary?
- When individual concentrations differ, what quantity must be averaged to obtain the population-average rate?
Solutions. (1) It is at least as large. The function \(1/(K_m+C)\) has second derivative \(2/(K_m+C)^3>0\), so Jensen’s inequality applies. (2) The individual rates are \(50(10)/(10+10)=25.0\ \text{mg/h}\) and \(150(10)/(30+10)=37.5\ \text{mg/h}\); their average is \(31.25\ \text{mg/h}\). The average parameters are \(V_{\max}=100\ \text{mg/h}\) and \(K_m=20\ \text{mg/L}\), giving \(100(10)/(20+10)=33.33\ \text{mg/h}\). (3) Varying \(V_{\max}\) introduces a product with \(1/(K_m+C)\), and dependence between the parameters affects the covariance; the fixed-\(V_{\max}\) inequality alone does not determine the result. (4) Average \(V_{\max,i}C_i/(K_{m,i}+C_i)\) over individuals, rather than evaluating the rate only at average values.