Sojourn Time (Epidemiology)

Sojourn time—also termed the preclinical detectable phase (PCDP) or detectable preclinical phase (DPP)—is the chronological interval in the natural history of a progressive disease during which a pathological lesion is biologically established and detectable by an asymptomatic screening assay, yet has not produced clinical signs or symptoms that would prompt standard medical diagnosis. Formalized mathematically by Marvin Zelen and Max Feinleib (1969) and expanded in multi-state Markov screening models by Stephen W. Duffy, Walter W. Hauck, and Nicholas E. Day, sojourn time represents the temporal window of opportunity within which population screening programs can intercept occult malignancies, metabolic disorders, or chronic infections. In quantitative epidemiology and biostatistics, the statistical distribution of sojourn times across a heterogeneous disease population governs four foundational parameters of public health screening policy: the maximum achievable lead-time bias, the magnitude of length-time bias, the incidence of interval cancers between scheduled examinations, and the optimal inter-screening interval required to reduce population mortality.

In chronic disease modeling, the trajectory of an untreated progressive disorder—such as colorectal adenocarcinoma, invasive breast carcinoma, cervical intraepithelial neoplasia, or abdominal aortic aneurysm—is partitioned into four sequential states separated by three transition boundaries:

1. Disease-Free State (S₀ → T₀): Prior to biological onset at chronological age T₀, the tissue lacks malignant transformation or pathological lesions, and any positive screening test result represents a false positive determined by the assay's specificity (Sp).
2. Occult Pre-Detectable Phase (T₀ to T_S): At age T₀, clonal initiation occurs, but the lesion remains below the physical or biochemical threshold of detection (for example, a solid tumor smaller than 1 to 2 mm in diameter or shedding insufficient circulating cell-free DNA). Between T₀ and T_S, existing screening modalities cannot distinguish the lesion from normal tissue.
3. Preclinical Detectable Phase / Sojourn Time (S = T_C − T_S): At age T_S, the lesion crosses the analytical threshold where a screening test with sensitivity Se can detect it in an asymptomatic individual. This preclinical sojourn persists until age T_C, when local invasion, vascular erosion, or mass effect triggers overt clinical symptoms (such as hemoptysis, palpable lymphadenopathy, or gastrointestinal bleeding) that lead to routine clinical diagnosis.
4. Clinical Disease Phase (T_C to T_D): Beginning at symptom onset T_C, the disease is clinically manifest and progresses until cure, remission, or disease-specific mortality at age T_D.

Within this four-state architecture, the sojourn time (S) for an individual case is defined by the difference S = T_C − T_S. If an asymptomatic individual undergoes a screening examination at chronological time T_screen within the interval [T_S, T_C) and tests positive, the resulting lead time (L)—the duration by which diagnosis is advanced ahead of clinical symptom onset—is L = T_C − T_screen. Consequently, for every screen-detected patient, the realized lead time is bounded above by that patient's sojourn time (0 < L ≤ S). Because T_S cannot be observed directly in any individual patient and T_C is censored once a screen-detected lesion is surgically excised, sojourn time is a latent random variable whose probability density function f_S(s) is estimated at the population level from longitudinal screening trial data.

Stationary Prevalence Pool and the Mean Sojourn Time Relation

When a previously unscreened population undergoes its initial prevalence screen (Round 1), the proportion of asymptomatic individuals harboring an occult lesion within the preclinical detectable phase depends on both the annual incidence rate (I) of clinical disease and the mean duration (E[S]) that lesions reside in the preclinical compartment. Under stationary epidemiological conditions where disease incidence and age structure are stable, the latent preclinical prevalence (P_preclinical) satisfies the balance equation P_preclinical = I × E[S]. If the screening test detects a fraction Se (test sensitivity) of all preclinical lesions present at the prevalence screen, the observed screen-detected prevalence yield (Y₁) at the first screening round is Y₁ = Se × I × E[S]. Rearranging this relationship yields the Prevalence-to-Incidence Ratio (PIR) estimator of mean sojourn time:

E[S] = Y₁ / (Se × I)

For example, if an unscreened population of women aged 50 to 59 years exhibits a baseline clinical breast cancer incidence rate of I = 2.5 cases per 1,000 person-years, and an initial mammography screening round with sensitivity Se = 0.85 detects Y₁ = 7.65 cancers per 1,000 women screened, the estimated mean sojourn time is E[S] = 7.65 / (0.85 × 2.5) = 3.6 years.

Length-Biased Sampling and Exponential Distribution Properties

In 1969, Marvin Zelen and Max Feinleib proved using stochastic renewal theory that cases intercepted by a cross-sectional prevalence screen do not have the same mean sojourn time as incident cases arising in the general population. Because a lesion remaining in the preclinical state for s years occupies a temporal target window proportional to s, the probability density function g(s) of sojourn times among screen-detected cases is weighted by length: g(s) = [ s · f_S(s) ] / E[S]. Taking the expectation of g(s) reveals the fundamental length-time bias identity:

E[S_screen] = E[S²] / E[S] = E[S] + [ Var(S) / E[S] ] = E[S] · (1 + CV_S²)

where Var(S) is the population variance of sojourn times and CV_S = σ_S / E[S] is the coefficient of variation. Whenever biological heterogeneity exists in tumor growth rates (Var(S) > 0), the mean sojourn time of screen-detected tumors (E[S_screen]) exceeds the population mean sojourn time (E[S]). In continuous-time Markov models of screening (Walter & Day, 1983; Duffy et al., 1995), the preclinical sojourn time S is modeled as an exponential distribution with transition hazard rate λ (where f_S(s) = λe^−λs, mean E[S] = 1/λ, and variance Var(S) = 1/λ², so CV_S = 1). Under this exponential distribution, length-biased sampling doubles the expected total sojourn time of prevalence-screened cases (E[S_screen] = 2/λ), while both the backward recurrence time and the forward recurrence time (expected lead-time bias E[L]) equal 1/λ.

Because the formula E[S] = Y₁ / (Se × I) contains two unknowns (Se and E[S] = 1/λ), joint identifiability is achieved by tracking interval cancers—clinical cases that surface symptomatically during the time interval t following a negative screening examination. Under the exponential distribution Markov model, an interval cancer arises either from a false-negative preclinical lesion missed with probability (1 − Se) that progresses to symptoms within time t, or from a de novo lesion entering the preclinical phase after the screen and traversing its sojourn window before time t. The expected cumulative incidence of interval cases I(t) approaches baseline incidence I according to I(t) / I ≈ 1 − Se · e^−λt. Fitting maximum-likelihood models simultaneously to prevalence yields (Y₁), incidence screen yields (Y₂, Y₃), and interval cancer rates provides separate estimates of Se and λ while avoiding verification bias and spectrum bias.

The empirical distribution of sojourn time governs how frequently a screening test must be repeated in a target population:

• Short Sojourn Time Disorders (E[S] < 2 years): When the preclinical phase is brief—such as breast cancer in women aged 40 to 49 years and non-small-cell lung carcinoma in heavy smokers—a 3-year interval allows most newly arising lesions to pass from T_S to T_C between visits, requiring annual screening in randomized controlled trials.
• Long Sojourn Time Precursors (E[S] = 8 to 15 years): When screening targets pre-invasive dysplastic lesions with decade-long sojourn windows (colonic adenomas or cervical intraepithelial neoplasia grade 3), screening every 3 to 10 years intercepts progressive lesions while limiting false-positive workups and incorporation bias.
• The Overdiagnosis Limit (T_S + S > T_D,competing): When the right tail of the sojourn distribution extends beyond remaining life expectancy from competing mortality causes, screening detects overdiagnosed pseudodisease that would never have caused clinical symptoms.

How does upgrading to a higher-sensitivity diagnostic technology affect sojourn time?

Sojourn time (S = T_C − T_S) depends on both the biological growth rate of the lesion and the detection limit of the screening assay. Replacing plain chest radiography with low-dose helical computed tomography (LDCT) shifts the threshold of preclinical detectability (T_S) earlier toward biological onset (T₀), lengthening the mean sojourn time and increasing the potential lead time.

Why is mean sojourn time shorter in younger patients for the same cancer type?

In mammographic screening trials, the mean sojourn time of breast carcinoma is 1.7 to 2.4 years in women aged 40 to 49 years compared with 3.5 to 4.2 years in women aged 50 to 69 years. Younger patients exhibit a higher proportion of rapidly proliferating, estrogen-receptor-negative tumors (advancing T_C earlier) and denser fibroglandular breast tissue that masks small masses on mammography (delaying T_S).

1. Zelen, M., & Feinleib, M. (1969). "On the theory of screening for chronic diseases." Biometrika, 56(3), 601–614. https://doi.org/10.1093/biomet/56.3.601
2. Walter, S. D., & Day, N. E. (1983). "Estimation of the duration of a pre-clinical disease state using screening data." American Journal of Epidemiology, 118(6), 865–886. https://doi.org/10.1093/oxfordjournals.aje.a113705
3. Duffy, S. W., Chen, H. H., Tabar, L., & Day, N. E. (1995). "Estimation of mean sojourn time in breast cancer screening using a Markov chain model of both entry to and exit from the preclinical detectable phase." Statistics in Medicine, 14(14), 1531–1543. https://doi.org/10.1002/sim.4780141404
4. Shen, Y., & Zelen, M. (2001). "Screening sensitivity and sojourn time from breast cancer early detection clinical trials: mammograms and physical examinations." Journal of Clinical Oncology, 19(15), 3490–3499. https://doi.org/10.1200/JCO.2001.19.15.3490