EVIDENCE LIBRARY
Can a Light Meal Make Breath Alcohol Peak Timing Too Variable for Straight-Line Back-Calculation? A 52-Person Study
A 52-person breath-alcohol study found peak times spanning 17–109 minutes. See what gamma modeling can—and cannot—prove in forensic BAC reconstruction.
Short answer: Yes, the timing of the breath-alcohol peak varied materially in this controlled experiment, and a straight declining line cannot represent the preceding rise, peak, and fall. Li and colleagues studied 52 adults in Hong Kong over 119 drinking sessions after a standardized light meal. The observed interval from the start of drinking to the peak breath-alcohol concentration ranged from 17 to 109 minutes. A nonlinear gamma-curve model fitted the complete measured breath-alcohol series better than competing statistical curve families under the study’s chosen model-fit criterion. Neither finding, however, establishes the breath or blood alcohol concentration of an individual defendant at an unmeasured earlier time.
The scientific basis is the complete original research article by Li, Sze, Wong, Tsui and So, published in PLOS ONE in 2019: Experimental study of the temporal profile of breath alcohol concentration in a Chinese population after a light meal (DOI: 10.1371/journal.pone.0221237). The important details are the actual breath measurements, the denominators behind the fitted curves, and the uncertainty created when an observation is moved backward in time.
What forensic question did the investigators test?
The investigators asked whether repeated breath tests after drinking could be described more appropriately by a curve that includes absorption and the peak, rather than solely by a constant-rate decline. This is relevant whenever a screening result and a later evidential result are separated by a meaningful time interval. A later, lower concentration is compatible with an earlier higher result if the individual was already on the descending limb. But it may also be compatible with an earlier lower result if the individual was still absorbing alcohol. Establishing which phase applied requires evidence beyond the two numbers alone.
The paper compared a zero-order linear model with nonlinear gamma, Weibull and lognormal specifications. These were statistical descriptions of a measured breath-alcohol time course; they were not independent analyses of venous whole blood and were not individualized forensic reconstruction validations.
Who participated, and what was actually measured?
The 52 recruited Chinese adults included 34 men and 18 women; their reported ages spanned 21–61 years. Participants underwent medical screening and were invited to attend two or three sessions separated by at least two days. The resulting 119 sessions comprised 83 male-participant sessions and 36 female-participant sessions. Multiple sessions from the same person are repeated observations, not 119 independent people.
Participants were instructed to avoid food for four hours and alcohol and sedatives for 24 hours before testing. They consumed a standardized light meal, and approximately 30 minutes later drank a 500 mL orange-juice-and-vodka preparation over 15–20 minutes. The study protocol allowed 0, 20, 40 or 60 g ethanol; the reported analytical session table covers 48 sessions at 20 g, 48 at 40 g and 23 at 60 g. Thus, the reported 119-session data should not be described as 119 placebo-controlled comparisons.
The physical test was an Alcotest 9510 evidential breath instrument using infrared and fuel-cell measurements. The researchers averaged its two sensor readings to obtain the study breath-alcohol concentration (BrAC). Measurements were scheduled roughly every 10 minutes during absorption and 30 minutes during elimination, with procedures lasting until a low breath threshold or eight hours. No venous blood draw or headspace-GC whole-blood ethanol assay is described as the analytical basis of this experiment. For the analytical distinction, see breath-alcohol measurement.
What did the directly observed breath data show?
Across the study sessions, the time from initial alcohol intake to the point treated as the start of the elimination phase ranged from 17 to 109 minutes. Recorded peak BrAC values ranged from 0.007 to 0.081 mg per 100 mL of breath. Those are ranges of measurements across sessions, not a confidence interval predicting one individual’s peak. The research does not provide a universal “wait exactly X minutes” rule or an individual probability of still absorbing alcohol at a particular time.
For the linear elimination analysis, researchers excluded BrAC readings below 0.009 mg/100 mL; the resulting linear fit drew on 1,492 breath records. Their more complete nonlinear analysis included 2,360 readings across the rise, peak and decline. Neither 1,492 nor 2,360 is a participant count. Treating serial tests from 52 people as thousands of independent case histories would exaggerate evidentiary precision.
| Item | Original result | Forensic meaning |
|---|---|---|
| Participants | 52 adults: 34 men, 18 women | Human subject count; not the number of serial assays. |
| Drinking sessions | 119 total: 48 at 20 g, 48 at 40 g, 23 at 60 g | Repeated visits contributed data; no 119-person independent cohort. |
| Direct sample and instrument | Exhaled breath, dual-sensor Alcotest 9510 | Not a laboratory whole-blood BAC measurement. |
| Observed peak BrAC | 0.007–0.081 mg/100 mL of breath | Observed session range; not an individual prediction interval. |
| Observed interval to peak/decline onset | 17–109 minutes from initial drinking | Broad observed timing; drinking and measurement schedules matter. |
| Serial breath records modeled | 1,492 linear-fit records; 2,360 nonlinear-fit records | Different inclusion rules and temporal coverage, not different subject cohorts. |
Why cannot a linear elimination line describe the whole drinking episode?
A descending zero-order line represents the assumption that alcohol concentration decreases by a constant amount per unit time once the relevant elimination phase is established. The preceding period is not a straight decline: gastrointestinal absorption and distribution can produce a rising curve, a peak and then a fall. Extending a downward slope backward through the ascending portion creates a concentration estimate that the line’s own assumptions do not justify.
The investigators estimated a group linear BrAC decline of 0.0042 mg/100 mL of breath per hour from the selected descending records. Their paper expressed that estimate as a calculated blood-equivalent decline of 9.7 mg/100 mL per hour using an assumed blood-to-breath conversion. That blood-equivalent value was not obtained by separately analyzing each participant’s whole blood. It cannot simply be substituted for a measured individual elimination rate in a litigation report.
Even an accurately measured breath decline in one subject must be tied to the relevant time interval, instrument procedures, sample quality and evidence of completed absorption. A historical concentration calculation additionally inherits uncertainty about actual dose, drinking duration, absorption phase, biological variation and the breath-to-blood relationship. See pharmacokinetics and toxicology for the distinction between concentration-time models and directly assayed specimens.
What made the gamma model different?
The investigators fitted right-skewed curves intended to represent the entire breath-alcohol trajectory. The gamma function has parameters governing curve range, shape and scale; the fitted forms allowed relevant parameters to depend on participant characteristics and the administered alcohol dose. Weibull and lognormal alternatives were evaluated as competing curve families. The group-level gamma model could rise to an estimated peak and descend, whereas a single descending line does not itself include the rising limb.
The preferred variable-parameter gamma specification had the lowest reported deviance information criterion (DIC) of 14,729, compared with 14,825 for the corresponding Weibull specification; the comparable lognormal version did not converge. DIC is a model-comparison statistic within the fitted data and assumptions. A lower DIC does not, without further validation, establish individual prediction accuracy, a forensic uncertainty interval, reliability on an unstudied population, or legal admissibility.
| Comparison | Reported value | Interpretation and limitation |
|---|---|---|
| Selected linear breath decline | 0.0042 mg/100 mL breath/hour | Estimated model slope for selected descending observations, not each participant’s measured metabolic constant. |
| Authors’ blood-equivalent expression | 9.7 mg/100 mL blood/hour | Conversion from breath using a presumed relation; whole blood was not sampled to validate it. |
| Gamma model, variable shape and scale | DIC 14,729 | Lowest reported DIC among assessed specifications, not a cross-validated case accuracy statistic. |
| Weibull model, variable shape and scale | DIC 14,825 | Higher DIC under the same comparison. |
| Illustrative modeled peak: 45 kg | 0.038 mg/100 mL breath | Model illustration for a 38-year-old man at 40 g alcohol; not a direct observation from a particular participant. |
| Illustrative modeled peak: 75 kg | 0.021 mg/100 mL breath | Same illustration, different assumed body mass; does not validate individual dosing history. |
Were the predicted peaks measured or simulated?
It is important to distinguish the observed ranges in Table 1 from the constructed examples in the original article’s Table 6. For a hypothetical 38-year-old man consuming 40 g of alcohol, the model yielded peak BrAC values of 0.038, 0.025, 0.023 and 0.021 mg/100 mL at modeled body masses of 45, 55, 65 and 75 kg. The corresponding modeled times to peak were 0.85, 0.91, 0.93 and 0.92 hours. These numbers show how changing an input affects a model, not what four identified people demonstrably experienced.
Those modeled results also illustrate why a single input such as body mass does not determine every feature of the concentration curve. A curve is conditional on its mathematical form, the assigned alcohol dose, the meal, its fitted parameters and the measurement process. The ability to generate a precise decimal is not the same as external experimental confirmation at that decimal.
Does the study prove a particular breath-to-blood ratio?
No. The researchers directly measured exhaled BrAC and referenced an assumed blood-to-breath relationship when expressing some breath estimates as blood equivalents. They did not collect contemporaneous paired venous whole-blood specimens to establish the person’s actual partition ratio or compare individually measured blood and breath kinetics. The quoted 9.7 mg/100 mL/hour conversion therefore cannot be presented as independently measured blood elimination.
Breath and whole blood also serve different regulatory and analytical purposes. A breath instrument result should be reported in the units and sample type actually analyzed. Any translation to a blood-equivalent concentration must disclose the adopted relation, its jurisdictional relevance and its uncertainty; it must never be relabeled as a measured venous whole-blood concentration. For specimen and testing distinctions, consult the forensic toxicology evidence library.
Can the 17–109-minute interval become an individual absorption deadline?
No. The lower and upper endpoints characterize the experiment’s observed session range under a particular drinking procedure after one standardized light meal. They do not form an empirically validated absolute minimum and maximum for all drinkers, body compositions, meals, gastric conditions, drinking histories or medications. A defendant’s record may involve different beverage strengths, meals, drinking duration, vomiting, absorption modifiers or delayed collection.
Equally important, an observed interval to a peak is not proof that absorption ended entirely at that timestamp. Circulating concentration reflects multiple simultaneous processes, and a peak can arise when net input and clearance temporarily balance. The peak is an observable turning point in a curve, not a direct gastrointestinal absorption assay.
What statistical and study-design limits matter in testimony?
First, the analysis came from 52 screened adults and repeated sessions. Serial breath observations within a participant and visit are correlated, and the model assumes a specified structure for the entire time course. Second, the gamma model was chosen on the reported DIC criterion; the study does not supply a prospective, independently held-out set of defendant-like cases demonstrating accurate reconstruction of a named person’s earlier concentration.
Third, the linear and nonlinear analyses included different sections of the time series and different numbers of records. A better whole-curve fit need not mean the same improvement at every legally important time point, especially near an uncertain peak. Fourth, the authors’ sex-, weight- and dose-related coefficients describe fitted associations within this dataset; they should not become categorical claims about a particular person’s metabolism or presumed legal impairment.
Finally, this was a pharmacokinetic breath-sampling experiment. It did not measure driving errors, standardized field sobriety performance, crash occurrence, arrest decisions or criminal causation. Statements about general traffic policy are distinct from evidence that a specific individual was impaired at a specific time.
How should an expert evaluate a delayed roadside and station breath-test comparison?
The forensic inquiry should begin by placing each actual measurement on a defensible time axis: the first and last drinks, meal timing, time of any driving or incident, each observed breath test, instrument type, analytical units, testing protocol and any possible contamination or sampling interruption. The model phase—ascending, near peak or descending—must be evaluated rather than assumed. An estimated earlier concentration should identify which observations are measured, which are inferred and which source-specific parameters were adopted.
In a case with sparse observations, different admissible kinetic histories can be consistent with the same late reading. The appropriate conclusion may be a range or an inability to reconstruct a unique value. The present study gives evidence of group variability after a light meal, not a validated correction factor that supplies missing timestamps, volumes, absorption state or specimen equivalence.
Frequently asked questions
1. Does this research show that everyone peaks 17–109 minutes after drinking?
No. That was the observed range across these experimental sessions under this protocol. It is not a universal human absorption window or a personal cutoff.
2. Did the investigators draw blood to measure BAC?
No. The source data for the reported profiles came from an evidential breath instrument. Blood-equivalent figures were calculated from breath values under an assumed conversion.
3. How many independent people were studied?
Fifty-two. They supplied 119 drinking sessions and thousands of serial breath readings. Neither sessions nor time points should be substituted for independent participants.
4. Did a lower DIC prove the gamma model predicts an individual defendant accurately?
No. It favored one modeled curve form under an in-sample comparison. It did not provide demonstrated individual forensic prediction error on independent casework.
5. Could a late breath result have been higher or lower earlier?
Yes. Depending on the phase, a lower late value may follow a higher peak, while an early rising phase can mean the earlier value was lower. The timing and evidence determine what can responsibly be inferred.
6. Does the paper establish an individual’s ethanol elimination rate?
No. The reported rate is an aggregate fitted model estimate, not repeated direct elimination measurements for the person in a new case.
7. Can the study establish impairment or responsibility for a crash?
No. The experiment measured breath-alcohol time courses and fitted statistical models, not actual driving outcomes or case-specific causation.
Bottom line for scientific evidence
After a standardized light meal, this study observed widely variable timing of peak breath alcohol across 119 sessions contributed by 52 adults. A gamma model described the full measured trajectory better than compared alternatives on the authors’ model-fit criterion; a simple declining line addresses only part of that trajectory. The study supports careful scrutiny of absorption phase, timestamps, input assumptions and breath-versus-blood distinctions. It does not justify assigning the study-average slope, a modeled peak or a presumed blood-equivalent BAC to a person without case-specific evidence and uncertainty analysis.
Primary source: Li YC, Sze NN, Wong SC, Tsui KL, So FL. Experimental study of the temporal profile of breath alcohol concentration in a Chinese population after a light meal. PLOS ONE. 2019;14(9):e0221237. doi:10.1371/journal.pone.0221237.