Why the Source Matters
A physiological value is not fully described by its number and unit. Its meaning also depends on who or what was measured, how the measurement was made, which endpoint was reported, and under what conditions. A value can be numerically plausible yet unsuitable for a particular PBPK reference profile because, for example, it comes from a different species, an unrepresentative population, or a method that measures a related but distinct endpoint.
The earlier material on reference physiological profiles established that a reference profile specifies the population and covariates to which model inputs are intended to apply. Source evaluation asks a narrower question: does this particular source provide evidence that supports an input for that specified profile? It is not enough to select the most recent table or the largest number of decimal places.
These labels describe the suitability of evidence for a specified use, not the general quality of a publication or database. A well-conducted study may be unsuitable for a different target population; a database may be convenient and carefully curated but still require checking its cited source and definitions.
A Practical Evidence Record
Before comparing values, write down the intended parameter and its target profile. Record the species, population characteristics relevant to the parameter, body-size basis, health status, and conditions such as rest, posture, or anesthesia when relevant. Also state the endpoint and its units. These details prevent a source from being judged against an unstated or shifting target.
For each candidate source, capture enough information to trace and interpret the value. A useful record includes:
- Provenance: the author or organization, publication or database version, date, and the primary study cited, if applicable.
- Population: species, age range, sex when reported or relevant, health status, and recruitment or inclusion criteria.
- Measurement: the method, setting, and relevant measurement conditions.
- Endpoint and basis: what was measured, whether it is a whole-organ or tissue-specific quantity, the units, and any normalization or conversion.
- Sample size: the number of individuals or experimental units contributing to the reported estimate, plus the number included in the relevant analysis if that differs.
- Summary and caveats: whether the value is a mean, median, range, or another summary, and any limitations stated by the source.
The word “sample size” needs care. A paper may recruit 30 people but report an endpoint for only 24; a paired measurement may yield observations from two organs per person, but that does not make the number of independent people 48. Record the denominator that actually contributed to the estimate, and clarify what that denominator counts. If the source does not say, mark it as unreported rather than guessing.
How the Source Type Affects Evaluation
| Source type | Strengths | Checks needed |
|---|---|---|
| Primary study | May provide direct details about recruitment, methods, conditions, and analyzed participants. | Inspect the methods and results; verify the endpoint, denominator, and applicability to the target profile. |
| Reference database | Can organize values consistently and make them easier to locate or compare. | Check definitions, version, provenance, and whether the entry is measured, derived, or transcribed from another source. |
| Secondary compilation | Can summarize a broad literature and provide useful context. | Trace the value to its underlying study where possible; check whether rounding, selection, or changed terminology obscures its basis. |
There is no automatic ranking in which one source type always wins. A primary paper may have strong methodological detail but a population that does not match the target. A database may have a useful standardized definition but provide insufficient detail in its short entry. A review may identify relevant studies but report a value without enough information to establish its denominator. Use the source type to guide the checks, not to replace them.
Reconstructing a Pooled Mean
When a source reports arithmetic means for separate subgroups, the overall arithmetic mean is a sample-size-weighted average, provided the subgroups are non-overlapping, their observations form the intended combined sample, and the reported subgroup means use the same endpoint and measurement basis. Let subgroup \(j\) have sample size \(n_j\) and mean \(\bar{x}_j\). Its observations sum to \(n_j\bar{x}_j\). Adding the subgroup totals and dividing by the total number of observations gives:
Thus, subgroup means alone generally do not determine the pooled mean: the weights are needed. There are special cases. If two subgroup means are identical, their pooled mean equals that shared mean regardless of their relative sample sizes. More generally, the available information may determine the pooled mean in some other special situation, but it should not be assumed to do so.
Worked Example: Combining Two Subgroup Means
A hypothetical study reports renal blood-flow means for two non-overlapping groups measured using the same method and conditions. Group 1 has \(n_1=8\) participants and mean \(0.90\) L/min; Group 2 has \(n_2=12\) participants and mean \(1.10\) L/min. Find the combined arithmetic mean for the 20 participants.
The subgroup means are in the same units and refer to the same endpoint, and the stated sample sizes count the participants contributing to those means. The pooled-mean formula therefore applies:
The combined mean is \(1.02\) L/min. If the subgroup sample sizes were not given, the mean would generally not be recoverable. For instance, weighting \(0.90\) and \(1.10\) L/min by sample sizes 12 and 8 gives \(0.98\) L/min, whereas equal sample sizes give \(1.00\) L/min. But if both subgroup means were \(1.00\) L/min, the pooled mean would be \(1.00\) L/min for any positive subgroup sizes.
Worked Source Appraisal
Worked Example: Checking Conditions Against a Reference Profile
Suppose a PBPK model needs a reference value for renal blood flow in healthy adults aged 20–40 measured at rest without anesthesia. A hypothetical primary study reports a mean of \(1.05\) L/min from 18 healthy adults aged 22–39, measured using MRI. The available description does not state whether participants were resting or whether any were anesthetized. Classify the evidence based on the information provided.
Population. The stated age range and health status align with the target profile. The reported denominator is 18 participants, so the estimate's sample size is identifiable from the description.
Endpoint and measurement. The reported endpoint is renal blood flow in L/min, and the measurement method is identified as MRI. The value is already in the requested unit. These details do not, however, establish that all relevant measurement conditions match.
Decision. This source is conditionally usable, not directly suitable on the stated information. Rest and non-anesthetized status are explicit requirements of the target profile, and the source description does not confirm them. The next step is to check the full methods or contact the source authors if appropriate. If those conditions cannot be confirmed, document the missing information; if a mismatch is established, do not present the value as directly representing the target profile.
Worked Example: Evaluating a Database Entry and Its Provenance
A hypothetical reference database lists renal blood flow as \(940\) mL/min and cites a primary study. The study reports a mean of \(0.94\) L/min for 16 healthy adults aged 25–35, measured at rest without anesthesia using the same endpoint. The target profile is healthy adults aged 20–40 at rest without anesthesia. The study states that 16 participants contributed to the mean.
First check the unit conversion: \(1\) L \(=1000\) mL, so \(940\) mL/min is \(940/1000=0.940\) L/min. This matches the primary study's \(0.94\) L/min after expressing the values to the same precision. The population age range, health status, endpoint, measurement condition, and method as described are consistent with the target; the analyzed sample size is also stated.
The cited primary study supports the database entry, so the value is traceable rather than an unexplained number. On the information given, the evidence is directly suitable for the stated target profile. Record both the database and the underlying study, along with the denominator and method. This conclusion does not mean that the value is universally appropriate for every adult population or every PBPK application.
Worked Example: Handling an Untraceable Compilation
A hypothetical secondary table lists “kidney perfusion: \(1.0\) L/min” for adults but gives no citation, age range, method, health status, measurement conditions, or sample size. A modeler wants a value for healthy adults aged 20–40 measured at rest without anesthesia.
The unit and broad organ label appear relevant, but “kidney perfusion” may not establish the exact endpoint required. The table does not show whether the number is a measured flow, an estimate derived from another quantity, or a rounded value taken from a different population. Nor can its denominator or study conditions be checked.
This entry is conditionally usable only as a lead for further searching, not as documented direct evidence for the model input. Look for a citation or a source with a definable endpoint and population. If the underlying evidence remains untraceable, do not fill in missing details by assumption; record the limitation and select a better-documented source for the specified use.
Diagnostics, Limitations, and a Common Pitfall
Several warning signs indicate that source evaluation is incomplete: a value lacks a cited origin; the reported endpoint has a different name or definition from the model input; the measurement basis is unclear; sample size is absent or ambiguously counts organs rather than participants; or important conditions are omitted. A unit conversion can resolve a units mismatch, but it cannot make different endpoints or populations equivalent. Likewise, rounding a number to fewer digits does not resolve uncertainty about its source.
A common mistake is to reason, “This is a peer-reviewed paper, so its value is directly suitable,” or “This is a curated database, so its number needs no further checking.” The correction is to separate source credibility from target suitability. Evaluate the actual population, method, endpoint, denominator, conditions, and provenance against the specified profile. When an essential detail is missing, use a conditional classification until the detail is verified; do not silently treat missing information as a match.
Source appraisal also has limits. Published reports may omit details, use historical definitions, or provide summaries that cannot be reconstructed from subgroup data. Even a fully traceable source may not represent every intended covariate or population. The task here is to state what the evidence supports and what it does not; a more detailed analysis of how to represent input uncertainty belongs to a separate step.
Check Your Understanding
Answer each question using the source-evaluation principles above. Solutions follow the questions.
- A study reports an organ-flow mean of \(0.80\) L/min for 10 participants and \(1.20\) L/min for 30 participants, using the same endpoint and conditions. What is the pooled mean?
- Two subgroup means are both \(0.75\) L/min, but their sample sizes are not reported. Can their combined mean be determined? Explain.
- A source matches the target population and reports the correct endpoint, but it does not state whether measurements were made under the target's required condition. How should it be classified based on the available information?
- A database reports a value with a citation, but the cited study measured a related endpoint rather than the model's requested endpoint. Does the citation alone establish suitability?
Solutions. (1) The pooled mean is \(\big[(10)(0.80)+(30)(1.20)\big]/40=(8+36)/40=1.10\) L/min. (2) Yes. Because both subgroup means are \(0.75\) L/min, every weighted average of them is \(0.75\) L/min, regardless of the positive subgroup sample sizes. (3) Classify it as conditionally usable pending confirmation; an unreported condition is not evidence that the target condition was met. (4) No. Provenance helps trace the value, but the endpoint mismatch must be resolved before the value can represent the requested input.