OBJECTIVE--The long-term predictability of left ventricular ejection fraction (LVEF) measurements was evaluated with particular emphasis on the assessment of deterioration in individual patients whose coronary artery disease was initially treated medically. DESIGN--Prospective pilot study. SETTING--Regional cardiac centre. PATIENTS--60 minimally symptomatic patients with coronary artery disease who, after arteriography, were initially treated medically. INTERVENTION--Measurement of LVEF by first pass exercise radionuclide ventriculography at baseline and six months later without interruption of usual anti-ischaemic medication. Baseline and six-month studies were analysed independently of each other. MAIN OUTCOME MEASURE--Based on 51 patients who remained event free, 95% prediction intervals were derived for prognostic LVEF indices to suggest the minimum change from baseline that might be considered clinically important in the individual patient, alerting clinicians to the need for closer review. RESULTS--At six-month ventriculography, 22 patients showed apparent deterioration in exercise LVEF or the change in LVEF with exercise (delta LVEF). Only two patients had six-month values below the lower limit of 95% prediction intervals, compared with 15 when 95% group confidence intervals (z = 3.33, p < 0.001) were used. When delta LVEF = 0 at baseline, the lower limit of 95% prediction intervals allowed for an exercise induced fall at six months of < or = 13%. For a baseline exercise LVEF of 50% (just normal), the lower limit of 95% prediction intervals was 38%--that is, the exercise LVEF could be measured as low as 38% > or = six months later without necessarily indicating or missing true deterioration. CONCLUSION--In the follow up of minimally symptomatic patients with coronary artery disease, serial long-term changes, in particular "deterioration," in prognostic LVEF indices may be interpreted more meaningfully with reference to 95% prediction intervals.