Back to Article:

How to predict benefits plan spending for January 2020

Add a comment

Have your say on this topic! Comments that are thought to be disrespectful or offensive may be removed by our Benefits Canada admins. Thanks!

* These fields are required.
Field required
Field required
Field required

Charles Spina:

Good article.

It’s a fairly straightforward actuarial exercise to predict period-to-period drug claim cost changes, and it’s certainly best consulting practice to devise reliable (i.e. best fit) formulas for this purpose, for which the author presumably has his own proprietary model.

Contrary to his assertion, though, historical data, especially for an 80,000 member plan population, is vital to those models’ predictive value. Only then can the relevant variables and their correlations be understood.

It would have been helpful if Mike had disclosed his reasons for predicting the cost reduction. Not saying it’s the case here, but the easiest back-of-the-napkin method of predicting cost changes with a reasonable degree of confidence is to make plan changes (e.g. co-pays, formulary deletions etc.) that have universal application to the member base. If that was what led to the cost reduction prediction, it isn’t highly sophisticated analytics.

Historical data is certainly needed to aid predictive modeling. Sometimes simple simulation using one of the drug claims payor’s “what if” applications (i.e. running last year’s data through the adjudication logic of pre/post plan changes) is all that is needed to calculate up a fairly reliable output value, before adjustments for regulatory and demographic changes. Having expertise with progressive or regressive med. consumption patterns for different disease states, which Mike has, would certainly be an advantage.

Tuesday, January 29 at 12:05 pm | Reply

Mike Sullivan:

Hi Charles – thanks for the thoughtful commentary. A few quick points: we are limited to 700 words, so it doesn’t allow for a deep dive.

I want to clarify something – yes you absolutely need historical data – but not for the trend, you need it to quantify disease state saturation relative to the demographic profile of the group, and you need it to determine disease severity profile. Those are critical to meaningful predictive analysis in health claims.

With respect to the case used – no the decline was not related to a plan design change or decrease in headcount. It was driven by a a few factors – one of which was enhanced COB with public plans (which can be forecasted if you track formulary status of claims and age/geography of claimants), and a change in mix of acute vs specialty claims. There are a few other contributors, but those are key aspects in this case.

The challenge for many in this area is that many common therapies can be used for multiple indications. If you have no way of isolating what a claimant is using a specific therapy for, and the severity of the condition, you can’t quantify disease prevalence and cannot match against demographics to determine future exposure and estimate timing.

Happy to chat about it in more depth. Don’t ever hesitate to reach out. Thanks again for the thoughtful questions and commentary. I hope your 2019 is off to a great start.

Thursday, January 31 at 3:12 pm

Join us on Twitter