QT Interval Measurements and Concentration-QTc Data Structure

Why the Data Structure Matters

Concentration-QTc analysis begins before a model is fitted: it begins with measurements whose meanings and timing are clear. An ECG provides an electrical recording from which intervals such as QT and RR are measured. A plasma sample provides a concentration associated with its own collection time. If the records are linked using only a planned visit label, or if replicate ECGs are collapsed without preserving their source values, the resulting exposure-response dataset can misrepresent what was measured and when.

This tutorial focuses on measurement and data organization. QT is the interval from the beginning of the QRS complex to the end of the T wave, measured in milliseconds. It reflects the time spanning ventricular depolarization and repolarization. RR is the interval between successive R-wave peaks, also usually measured in milliseconds; it describes the length of a cardiac cycle. The two are related physiologically, but they are distinct recorded quantities. QT correction and baseline-adjusted QTc are separate topics.

From ECG Traces to Replicate Summaries

A measurement record should identify the participant, study, treatment or period, visit, planned time, actual ECG time, replicate number, QT value, RR value, measurement units, and relevant quality or reader information. The original replicate-level observations should be retained even when a summary is created. They allow review of outlying values, missing replicates, and data processing decisions.

For a set of \(m\) valid replicate QT measurements collected in a defined measurement window, the arithmetic mean is

The corresponding mean RR is calculated from the RR values in that same replicate set:

Keeping QT and RR paired by replicate is important. If replicate 1 has a QT and RR recorded from the same ECG, those values form a measurement pair. Averaging each column separately produces a useful summary of the set, but it does not mean that every individual QT value was observed at the average RR. Do not create a new QT-RR pair by combining measurements from different people, visits, or unrelated times.

A mean is not automatically appropriate for every dataset. First check the protocol and data conventions: how many recordings constitute a set, what makes a replicate invalid, and how missing or unreadable traces are handled. Report the number of contributing replicates alongside a summary. If only two of three scheduled recordings are usable, the mean of those two is not a three-replicate mean. Do not silently substitute a different summary rule or exclude a value merely because it looks inconvenient.

Why Replicates Can Help—and What They Cannot Fix

Replicate measurements can reduce the effect of random measurement error when the errors are independent, centered around zero, and have the same variance. To see why, let a replicate measurement equal a common underlying value plus an error: \(Y_j=\theta+\epsilon_j\), where \(E(\epsilon_j)=0\) and \(\operatorname{Var}(\epsilon_j)=\sigma^2\). If the replicate errors are independent, then

Thus, under those assumptions, averaging \(m\) replicates reduces the error variance compared with relying on one replicate. Independence can be unrealistic when recordings are close together or share reader and instrument effects. If each error has variance \(\sigma^2\) and every pair has correlation \(\rho\), then

When \(\rho>0\), the variance reduction is smaller than the independent-error calculation suggests. Averaging also does not remove systematic bias, a shared technical problem, or genuine changes over the measurement window. Replicates are most interpretable when they are collected under a protocol-defined timing rule and the individual results are inspected for quality.

Nominal Time, Actual Time, and Concentration Matching

A nominal time is a design label, such as “2 hours after dose.” It is useful for scheduling and describing the study design, but it is not proof that a measurement occurred exactly two hours after dose. An ECG may be delayed, and a blood draw may occur earlier or later than planned. Record the actual ECG time and actual plasma collection time, with a consistent reference such as elapsed time from dose. Store clock timestamps and time-zone or date information when needed to resolve the sequence unambiguously.

A concentration is associated with the plasma sample’s collection time. An ECG is associated with the time its recording was obtained. Linking them for exposure-response analysis requires more than a shared nominal visit label. At minimum, verify that the participant, treatment or period, dose occasion, and intended assessment window agree. Then assess and record the difference between actual collection times. A concentration drawn at a different time is not literally a contemporaneous measurement of concentration at the ECG time.

A Practical Record Layout and Matching Workflow

A useful structure distinguishes the ECG replicate table from the concentration table, then creates an analysis table with an explicit link between the relevant ECG summary and concentration record. An ECG table can have one row per replicate, including participant ID, treatment or period, nominal time, actual ECG time, replicate number, QT, RR, units, and validity status. A concentration table can have one row per plasma sample, including participant ID, treatment or period, nominal time, actual draw time, concentration, units, and sample status. A summary or analysis table should preserve identifiers for the contributing ECG replicates and the linked plasma sample.

The key is to avoid making the data structure hide the timing information. If one plasma sample is associated by protocol with a triplicate ECG set, it may be linked to that set’s summary while still retaining the three ECG rows and the plasma sample’s exact collection time. The data should make clear that the concentration is a single sample associated with a set, not three independently collected concentrations. Similarly, repeated ECG sets from the same participant are repeated observations, not separate participants.

Diagnostics, Alternatives, and Common Pitfalls

Before using a concentration-QTc dataset, check that QT and RR are expressed in consistent units, timestamps are interpretable, replicate counts agree with the protocol, and summary values can be reproduced from source records. Investigate missing or duplicate replicate numbers, impossible or reversed time sequences, multiple candidate plasma samples, and changes in units or identifiers. Review ECG quality flags and the source trace when a value is unusual; a numerical outlier may reflect either a genuine measurement or a recording, reading, or data-entry problem. Do not impose an arbitrary exclusion threshold without a scientific and protocol-based reason.

Two common data strategies are to analyze replicate-level ECG records or to summarize replicates at a planned assessment occasion. Retaining replicate-level records preserves detail and supports review, but it requires later analysis to account appropriately for repeated records. Summarizing produces a compact occasion-level dataset and is straightforward when the protocol defines a replicate summary, but it can conceal within-set variation or missingness. Whichever strategy is used, preserve enough source information to reconstruct the other view.

For concentrations, a directly measured sample close in time to the ECG differs from a concentration estimated at the ECG time. The first is an observation at its documented collection time; the second depends on an estimation method and its assumptions. A nominal-time label is neither of these. This tutorial’s matching rules ensure that the distinction remains visible; they do not prescribe a method for estimating an unobserved concentration.

• Mistake: treating nominal time as actual time. “Both were scheduled at 2 hours” does not show that the ECG and sample were collected together. Use actual timestamps and the protocol’s matching rule.
• Mistake: selecting the nearest concentration regardless of context. A nearby sample from another treatment period or dose occasion is not a valid match. Check participant and occasion identifiers first.
• Mistake: averaging without preserving replicate records. A summary alone prevents review of replicate quality and calculation. Keep source values, validity status, and the number of contributing ECGs.
• Mistake: treating each replicate as an independent participant-level observation. Replicates belong to a measurement set from the same participant and occasion. Their relationship must remain identifiable in the data.