Reference range

A reference range is an interval derived from measurements in a specified reference population under specified analytical conditions. It gives context for a laboratory result; it is not automatically a boundary separating disease from health. Different methods, ages and reference populations can lead to different intervals. ^[ref]

For an illustrative central 95% interval, about 5% of measurements from the defining population lie outside by construction. If 20 independent tests each have a 5% outside probability, the probability of at least one outside result is $1-0.95^{20}\approx 0.64$. Independence is an assumption, but the calculation shows why many tests increase incidental flags. A clinical decision threshold can instead be chosen for a particular diagnostic or treatment purpose.

Population conditioning of ventricular reference ranges describes how the comparison interval for a ventricular measurement depends on the population and measurement conditions used to define it. A reference interval for a specified age, body size and imaging protocol is a different statistical object from an interval pooled across an arbitrary mixture of groups.

Let $Y$ be a ventricular measurement and $Z$ the recorded conditioning variables. A conditional central $95\%$ interval iswhere $Q_p$ is a quantile. It describes a reference distribution under those conditions; it is not automatically a disease-decision boundary or a confidence interval for a patient's true value.

A primary cardiac-MRI study used 120 healthy participants, 60 men and 60 women aged 20–80, with steady-state free-precession cine imaging and three-dimensional right-ventricular modelling. Age, body surface area and sex predicted several measurements; some age associations remained after size normalization.^[ref] Dividing a volume by body surface area therefore need not remove every population association.

An original example separates conditioning from pooling. Suppose hypothetical groups have volume distributions $N(80,10^2)$ and $N(120,10^2)$ in millilitres. Their central normal intervals are approximately $60.4$–$99.6$ and $100.4$–$139.6\mathrm{mL}$. A value $110\mathrm{mL}$ is three standard deviations above the first group's mean but one below the second's. The groups are mathematical examples, not assigned clinical categories or published thresholds.

For a population mixture, $F_{\mathrm{pool}}(y)={\sum }_g{\pi }_g{F}_g(y)$. Its quantiles depend on the mixing proportions ${\pi }_g$ and generally are not averages of group quantiles. Changing the sampled proportions can change the pooled reference interval even if every within-group distribution stays fixed.

A transferable reference description therefore states participant selection, conditioning variables, acquisition method and segmentation conventions. Associations with recorded categories do not by themselves identify causal biological mechanisms. Interpretation additionally needs measurement uncertainty and the clinical question; membership outside a reference interval alone does not establish a diagnosis.