Well, estimations are close anyway...
Money well spent. /s
Confirmation carries a cost and I'm sure they're not running tests on everybody who has a singular symptom that MIGHT indicate flu.
I'm guessing too that they might have models based on a number of factors that allow them to produce a rough estimate based on statistical elements such as severity spread of cases, actual confirmations vs referrals, and other elements.
Consider this extreme simplification: IF you knew that the severity distribution followed a bell curve, and all you had was some of the upper region of that curve, you could fairly reasonably extrapolate the rest of the curve.