Objectives Over diagnosis will occur if those with other diseases are included within the target’s diagnostic criteria and also if a criterion’s test result threshold is set too low so that those without the ‘disease’ are included. Borderline test results will contain a lower proportion of those benefiting whereas more ‘severe’ test results will contain a higher proportion that benefit. In order to differentiate between these heterogeneous groups, test results that reflect disease severity have to be used and analyzed numerically as opposed to results being dichotomized into positive and negative and analyzed using the indices of sensitivity and specificity. The object of this work was to create curves that show the probabilities of an outcome on treatment and control (with confidence interval for differences) conditional on different degrees of disease severity in order to establish thresholds that avoid over-diagnosis and over-treatment.

Method The methods were developed using data from screening diabetic patients for Albuminuria. Differential diagnoses leading to over-diagnosis of Albuminuria (e.g. UTI, etc) were excluded with relevant results (e.g. a negative MSU, etc). The effects of hypertension and Diabetes Mellitus were minimized by strict control of the latter. The current provisional threshold for ‘confirming’ a diagnosis of Albuminuria is an albumin excretion ratio (AER) of 20mcg/min (i.e. the upper 1.96 SD of the natural log of AER in the general population). Those above the AER threshold were randomised to placebo or irbesartan and those with an AER above 200mcg/min after two years labelled as having the surrogate biochemical outcome of ‘Nephropathy’. The frequencies of nephropathy in different AER ranges on placebo and irbesartan were used to create a calibrated pair of logistic regression curves with 95% confidence limits, odds ratios and risk ratios for each value of AER.

Results The curves displaying the probability of nephropathy on placebo and irbesartan for each value of AER were sigmoid in shape. The estimated probability of nephropathy conditional on an AER of 20mcg/min was 0.044 for placebo and 0.030 for treatment (NNT=77). The estimated probability of nephropathy conditional on an AER of 200mcg/min was 0.757 for placebo and 0.496 for treatment (NNT = 3.3). The average probability of nephropathy for placebo was 0.153 and for treatment was 0.077 (difference of 0.076, NNT = 13 and 95% CI 0.133 to 0.0185). The lower 95% CL for the risk difference did not exceed zero until the AER was over 35mcg/min when the risk difference was 0.021 (95% CI 0 to 0.042). At an AER of 200mcg/min the difference was 0.260 (95% CI was 0.17 to 0.35). Thirty-six percent (204/571) of study patients had with an AER between 20 and 40mcg/min.

Conclusions Instead of being restricted to the average outcome of a RCT, the decision maker can see the outcome probabilities at each value of the AER together with their differences and the CI of the former and latter. When creating a guideline, it is thus possible to choose a threshold that would correspond to the AER below which very few accept treatment when offered it during shared decision making. This might be an AER of 40mcg/min on the basis of the above results (if so, this would mean 36% of those previously diagnosed with Albuminuria were over-diagnosed). When this information is applied to multiple risk factors that also contribute to treatment heterogeneity, it is important to make careful causal inferences. For example irbesartan should cause the BP to reduce as well as the AER, both thus reducing risk of nephropathy but irbesartan does not cause reduced risk via blood glucose reduction.