Some of our clients struggle with “incidence” and “prevalence”, specifically their application in budget impact analysis (set up of a patient flow). Based on our daily analyses, we are happy to share our insights.

To start, both concepts must be well understood. We will illustrate them using patients with a specific condition, for example, patients with prostate cancer.

Incidence refers to the number of newly diagnosed patients within a specific time frame, such as the number of patients with newly diagnosed prostate cancer within one year. Incidence can be expressed as a percentage relative to the target population (e.g., male adults in Belgium).

Cumulative incidence refers to the number of patients that can be counted over time starting from a specific point in time. If this accumulation occurs over a longer period, it may exceed prevalence since no correction is made for dropout and/or mortality.

Prevalence refers to the total number of still-living patients who meet a specific criterion (e.g. prostate cancer) at a given point in time. Theoretically, you can snap them in 1 picture. The number of prevalent patients can be lower or higher than the incidence over a given period (dependent of the mortality and/or health condition in your analysis).

Let’s start with a diagram. The diagram projects the cumulative incidence over four years (a hypothetical example) as well as the prevalence, expressed as a percentage. If the timeline were 10 years, then the entire pool of prevalent patients would be “replaced” by ‘de novo’ patients starting from time point 0. In reality, this period may be slightly longer than 10 years because some patients from the newly diagnosed pool will also pass away.

Internationally developed budget impact models often calculate based on cumulative incidence while also partially incorporating the prevalent pool.

Moreover, some budget impact models calculate cumulative incidence ‘on top’ of previous years. This overestimation is not (or insufficiently) corrected for dropout (i.e., the green area is added on top of the blue area).

In reality, of course, you can never have more than 100% prevalent patients.

Developers inform us that this ‘overestimation’ is not an issue because a budget impact analysis compares two scenarios, each generating the same ‘overshoot.’ As a result, these effects cancel each other out. However, this is only valid if both scenarios are identical (same specialties and costs). In our opinion, a budget impact analysis then becomes meaningless (since there would be no difference between the two scenarios).

Another justification for combining prevalence and incidence is to account for long-living patients in the analysis. It may take a long time for a diagnosed patient to be eligible for the analysis; otherwise, these patients would not be accounted for.

This is partially true, but including patients who were diagnosed several years ago is also indirectly accounted for based on a 100% incidence model.

Both approaches (prevalence and/or incidence models) have one major shortcoming: they do not adequately account for changes in innovation (over recent years) and optimized diagnostic techniques. In other words, a patient diagnosed 10 years ago may have a lower survival chance than a similar patient diagnosed today. The treatment pathway may now be different, and the diagnosis may have been optimized due to innovation. In a health economic model, adjustments are made for this (discounting). In a budget impact model, this is not done, and certainly not retroactively.

A model may therefore contain potentially two shortcomings: incidence ‘on top’ of prevalence, as well as insufficiently correction for changes in clinical practice due to innovation.

These are, of course, simulations, and as long as the (financial) impact is limited (and thus the decision potentially influenced by it), this uncertainty can be included in a sensitivity analysis. However, for slow-progressing diseases, it is best to consult clinical experts to validate the numbers and proportions used in the analysis (patient flow).

Nevertheless, there are undoubtedly many opinions on this matter.

In a simplified format (as in the top figure), one can assume that the inflow of (incidence) patients is stable. Over time, a ‘steady state’ occurs (inflow (green area) = outflow (also green area)), and the total number of patients (upper boundary of the blue area) represents the ‘prevalent population.’

The volume of patients within a prevalent pool generally remains relatively stable unless there are significant innovations and/or demographic shifts and/or epidemiological events (e.g., COVID-19).

The above diagram does not show an increase/decrease in prevalence. In practice, this will evolve slightly, which may have a limited impact on a budget impact analysis.

For simplicity, we exclude optimized diagnostic techniques and changes in the therapeutic landscape or incidence trends.

Is it possible to develop a qualitative budget impact analysis solely based on incidence?

We believe this is absolutely possible.

Based on the above, one can reason that all patients who ever enter the simulation model are ‘incident.’ Therefore, an incidence pool can be used where each patient will eventually transition to another ‘health state’ (including death) over time. The strict definition of ‘incidence’ is thus somewhat unfortunate in terms of terminology. Therefore, it is best to distinguish between ‘de novo’ (first diagnosed, regardless of disease stage) versus ‘incidence at the treatment line level’ (e.g., advanced disease stage, grade III+). For slowly progressing patients, it may take years before they enter a specific treatment line. However, this effect balances out (some patients were diagnosed long ago, and newcomers (‘de novo’) may wait sometime over several years (‘watch and wait’) before qualifying for treatment). These numbers ‘cancel each other out,’ so there is no need to consider the prevalent pool.

In such an analysis, we refer to ‘number of patients’ rather than ‘patient years.’

The key advantage of an incidence model is that it can be linked to treatment duration. Each patient is assigned a specialty (or treatment) they receive over a number of cycles.

The sum of all treatment months (for all patients, including ‘watch and wait’) theoretically corresponds to the total number of months survived by all patients. Dividing this by 12 gives the theoretical number of treatment years.

This number should precisely match the prevalence pool (averaged over one year), also referred to as ‘patient-years’.

Is it possible to calculate solely based on the prevalence pool without considering incidence?

Yes, this is also possible.

Then, a treatment distribution must be projected onto the prevalent pool without considering treatment duration. Market shares must therefore be estimated ‘intuitively’ and logically.

When estimating market shares, it is crucial to determine whether the projections are made concerning incidence or prevalence patients.

Which is preferred: incidence or prevalence?

Based on our expertise, we recommend using incident based models for periodic treatments where the patient does not return to a previous health state. This applies to all oncology, including slow-progressing cancers such as leukemia.

Prevalent models are ideal for chronic conditions where patients are treated continuously or temporary and may return to a previous health state. Treatment of migraine is a good example.

How to handle models that calculate using both prevalence and incidence?

We recommend correcting such simulations for ‘outflow’ (mortality). In other words, one of the (selected) comparators should be ‘off treatment’ (including death).

Ultimately, it is essential to have a clear understanding of incidence and the likelihood of these patients qualifying for treatment. This forms the basis for potential subsequent treatment lines.

If you would like more insight, feel free to contact our expert team.