Why Overnight Glucose Needs a Time-Varying Model
A fasting glucose concentration is not necessarily stationary simply because no meal is being absorbed. Hepatic glucose output, insulin action, and counterregulatory hormone concentrations can vary across the night. Their combined effects may change the balance between glucose entering and leaving the circulating pool, producing an overnight pattern or a rise toward morning.
A model should distinguish the clock-time pattern from the mechanisms represented by that pattern. A periodic term can describe a reproducible daily rhythm, while measured insulin, glucagon, cortisol, or other hormone profiles can represent candidate mediators. These are not interchangeable explanations: a time-of-day association alone does not establish which biological pathway caused it.
A Mechanistic Glucose-Balance Model
Let \(G(t)\) be plasma glucose concentration in mmol/L at clock time \(t\), measured in hours. For a fasting interval, a useful mass-balance structure is
Here \(V_G\) is an apparent glucose distribution volume in L/kg body weight, and the fluxes on the right are expressed in mmol/kg/h. \(EGP(t)\) is endogenous glucose production, predominantly hepatic in this fasting context. The removal terms represent insulin-independent disposal, insulin-mediated disposal, and urinary glucose loss. If the model is not intended to represent renal loss, the final term may be omitted only when that omission is appropriate for the glucose range being modeled.
The equation states that concentration rises when total input exceeds total removal. In the earlier material on renal glucose filtration and SGLT-mediated reabsorption, renal loss is treated as a glucose-dependent process; here it is included as one term in the overnight balance rather than developed as a separate renal model.
One way to represent delayed insulin action is with a dimensionless effect state \(A_I(t)\):
\(I_n(t)\) is insulin normalized to a chosen reference concentration, \(S_H(t)\) is relative hepatic or systemic insulin sensitivity, \(k_A\) has units h\(^{-1}\), and \(CL_I\) has units L/kg/h. This simple effect compartment assumes insulin action approaches its current target \(S_H I_n\) with a first-order delay. It is an approximation, not a claim that all insulin actions share one time constant.
A daily rhythm in sensitivity can be represented by a periodic function:
The period is 24 hours, \(\phi_S\) is the time of the maximum, and \(a_S\) is the fractional amplitude. The bound on \(a_S\) keeps sensitivity positive. An analogous term can represent an intrinsic rhythmic component of hepatic output. Explicit hormone effects can also be included, for example as a function of glucagon and cortisol concentrations. If both a direct time-of-day term and hormone-specific terms are fitted, the direct term should represent residual rhythmic variation rather than re-counting the same hormone-mediated effect.
Assumptions and a Short Derivation
The balance equation assumes the modeled pool is adequately represented by one effective concentration, fluxes are expressed on a consistent body-weight basis, and the fasting interval has negligible meal-derived glucose input. It also assumes the specified flux functions describe the relevant glucose range. In a person with appreciable glycosuria, for example, setting renal loss to zero may materially distort the predicted trajectory.
For interpretation and quick calculations, the full model can be approximated locally near a reference concentration \(G_{\mathrm{ref}}\). If the net glucose flux is linearized around that point, the resulting reduced model can be written
Here \(q(t)\) is the net imbalance at \(G_{\mathrm{ref}}\), in mmol/L/h, and \(k\) is a local feedback or clearance rate in h\(^{-1}\). To derive its solution over an interval with constant \(q\), let \(D(t)=G(t)-G_{\mathrm{ref}}\). Then \(dD/dt=q-kD\). Multiplying by \(e^{kt}\) gives \(d(e^{kt}D)/dt=qe^{kt}\). Integrating from \(0\) to \(t\) yields
This solution is valid only while \(q\) and \(k\) can reasonably be treated as constant. For time-varying inputs, the model must be integrated across the changing inputs, numerically or piecewise. The reduced equation is useful for understanding local dynamics; it does not replace a mechanistic model when individual fluxes are needed.
Worked Example: Calculating a Rhythmic Hepatic-Output Component
Suppose a hypothetical model represents the intrinsic rhythmic component of fasting hepatic output as \(EGP_{\mathrm{circ}}(t)=E_0[1+a_E\cos(2\pi(t-\phi_E)/24)]\), with \(E_0=2.00\) mmol/kg/h, \(a_E=0.15\), and \(\phi_E=6\) h, where \(t=0\) is midnight. Hold hormone concentrations and insulin action fixed to isolate this component.
At 02:00, \(t=2\), so the cosine argument is \(2\pi(2-6)/24=-\pi/3\), whose cosine is \(0.5\). Thus
At 06:00 the cosine is 1, giving \(EGP_{\mathrm{circ}}(6)=2.00[1+0.15]=2.30\) mmol/kg/h. At 22:00 the cosine is \(-0.5\), giving \(EGP_{\mathrm{circ}}(22)=2.00[1-0.075]=1.85\) mmol/kg/h. These are values of the specified rhythmic component under fixed conditions, not complete predictions of hepatic output or of glucose concentration.
From a Net Imbalance to an Overnight Trajectory
Glucose concentration reflects the accumulated consequences of flux imbalance, not the value of hepatic output alone. Even if output rises, glucose need not rise if disposal increases by the same amount. Conversely, a modest sustained imbalance can change concentration over several hours. The following reduced-model calculation illustrates how both the duration and size of an imbalance matter.
Worked Example: A Two-Phase Overnight Glucose Rise
Consider a hypothetical fasting trajectory from 22:00 to 06:00 with \(G_{\mathrm{ref}}=5.50\) mmol/L and \(k=0.40\) h\(^{-1}\). Let \(G(22{:}00)=5.50\) mmol/L. Assume \(q=0.050\) mmol/L/h for the first six hours, followed by \(q=0.250\) mmol/L/h for the final two hours. These \(q\) values summarize the net imbalance near \(G_{\mathrm{ref}}\); they are not hepatic-output fluxes.
For the first six hours, the solution gives
Therefore \(G(04{:}00)\approx 5.614\) mmol/L. For the next two hours, the starting deviation is approximately 0.1137 mmol/L, so
The predicted 06:00 concentration is approximately 5.895 mmol/L. This example demonstrates how an increased late-night imbalance can produce a dawn rise even though the model includes a restoring term. It does not show that a particular hormone or pathway caused the rise: that requires a model with identifiable pathway inputs and suitable observations.
Separating Candidate Contributions
A useful mechanistic analysis asks how the predicted balance changes when one modeled driver is altered while other inputs are held fixed. Such counterfactual calculations are conditional on the model structure and parameter values. They quantify modeled contributions; they do not by themselves establish causal effects in a patient.
Worked Example: Comparing Output and Sensitivity Contributions
Suppose a hypothetical flux table gives nighttime hepatic output of 2.00 mmol/kg/h and total glucose removal of 1.98 mmol/kg/h. Near dawn, output is 2.06 mmol/kg/h while removal is 1.95 mmol/kg/h, because the modeled insulin-mediated component of removal is lower. Let \(V_G=0.20\) L/kg, and use these instantaneous fluxes to compare rates at the two times.
At night the net input is \(2.00-1.98=0.020\) mmol/kg/h, so the instantaneous concentration rate is \(0.020/0.20=0.100\) mmol/L/h. Near dawn it is \(2.06-1.95=0.110\) mmol/kg/h, giving \(0.110/0.20=0.550\) mmol/L/h. The dawn rate is therefore higher by \(0.450\) mmol/L/h at these specified states.
For a one-factor comparison, retain nighttime removal while changing only output: the net flux becomes \(2.06-1.98=0.080\) mmol/kg/h, an increase of 0.060 mmol/kg/h from the nighttime balance. Retain nighttime output while changing only removal: the net becomes \(2.00-1.95=0.050\) mmol/kg/h, an increase of 0.030 mmol/kg/h. Together these changes account for the observed net-flux difference of \(0.060+0.030=0.090\) mmol/kg/h, or \(0.090/0.20=0.450\) mmol/L/h.
This arithmetic separates two contributions for this additive flux balance. In a nonlinear model, interactions can make the allocation depend on the order of the counterfactual changes. The instantaneous rates also should not be extrapolated unchanged for hours: glucose-dependent flux feedback and changing hormones may alter them.
Diagnostics, Alternatives, and Limitations
A fitted rhythmic model should be checked against the observed time course, not just against a dawn-versus-night average. Plot residuals by clock time and inspect whether the model systematically underpredicts or overpredicts at particular phases. Check that predicted glucose and all fluxes remain physiologically admissible, that the rhythmic amplitude does not imply negative sensitivity or output, and that the phase and amplitude are supported by repeated measurements rather than a single noisy profile.
A 24-hour harmonic is a compact description of a smooth daily pattern, but it may miss abrupt sleep-related changes or multiple peaks. A piecewise model can represent a transition at a specified time, but its predicted change may depend strongly on the chosen boundary. A model with measured hormone profiles can test whether their timing is consistent with glucose changes, but hormone concentrations alone may not identify their effects separately from correlated insulin action or other time-varying inputs.
| Approach | Useful when | Main limitation |
|---|---|---|
| Constant fasting parameters | A short interval has no clear time-of-day pattern | Cannot represent a systematic overnight rhythm |
| Periodic time-of-day terms | Repeated observations support a smooth daily pattern | Associations with clock time do not identify biological mediators |
| Explicit hormone and insulin-action inputs | Relevant profiles are measured and mechanisms are the question | Correlated inputs and sparse sampling can make effects difficult to distinguish |
| Piecewise time-varying inputs | A transition is expected at a known time | Results can be sensitive to the selected transition time and interval structure |
A major design limitation is confounding between circadian timing and behavior. Clock time may track sleep, posture, activity, light exposure, or medication schedules. Repeated sampling across nights, careful recording of those conditions, and measurements of candidate mediators strengthen interpretation. Even then, a model may not uniquely identify mechanisms if different combinations of output and disposal produce similar glucose trajectories.
The main practical pitfall is to interpret a morning increase as proof that hepatic production alone increased. The glucose balance depends on all relevant inputs and removals. A rise can reflect greater output, reduced insulin-mediated disposal, changes in counterregulatory drive, or combinations of these. The correction is to write the balance explicitly, inspect the time courses of its components, and state which effects are measured and which are assumed.
Check Your Understanding
Answer each question, then compare with the brief solutions below.
- In a fasting glucose-balance model, what does a positive net flux imply for \(dG/dt\), assuming \(V_G>0\)?
- A periodic sensitivity model has \(a_S=0.20\). What are its maximum and minimum values relative to \(S_0\)?
- In the reduced model, let \(G(0)=G_{\mathrm{ref}}\), \(q=0.10\) mmol/L/h, \(k=0.50\) h\(^{-1}\), and \(t=2\) h. Calculate \(G(t)-G_{\mathrm{ref}}\).
- Why does a fitted clock-time rhythm not, by itself, show that cortisol caused a dawn rise?
Solutions. (1) Since \(V_GdG/dt\) equals net flux and \(V_G>0\), a positive net flux implies \(dG/dt>0\). (2) The maximum is \(S_0(1+0.20)=1.20S_0\), and the minimum is \(S_0(1-0.20)=0.80S_0\). (3) The deviation is \((0.10/0.50)(1-e^{-0.50(2)})=0.20(1-e^{-1})\approx0.1264\) mmol/L. (4) Clock time may covary with other rhythms and behaviors; the fitted term describes timing, not a uniquely identified mediator. A cortisol mechanism requires appropriate hormone measurements and a model that can distinguish its effect from competing inputs.