Prevalence at 10% of what is should be.

[jackolney]

Why?

• [x] Ensure that Scheduled Incident cases actually happen
• [x] Remove distribution of cases throughout a year (test)
• [x] Don't randomise cases (test)
• [x] Why are incidentCases not matching those that are scheduled to occur? (they are but incidentCases is 1 year behind)
• [x] Check the timings of scheduling ( using previous years incidentCases to inform the next years)
• [x] How does PLWHIV compare with Spectrum? (may have to do 15to49)

[jackolney]

Beta should be included post-2002 Priority is to fix pre-2002 first

[jackolney]

I've got a feeling that IRR are being picked from the beginning of the distribution only. i.e. from early ages only.

• [x] Reverse the distribution (run for loop in reverse) and see if effect is flipped.

[jackolney]

I think this is now an issue with rounding in population::CalculateIncidence()

        /* Find Incidence(a,s) */
        for(size_t j=0;j<34;j++) 
            incidence[j] = Round(lambda * people.at(j).size() * IRR[j]);

Because we Round() the Susceptibles[j] * IRR[j] for each Age and Sex vector then we lose some resolution in our incidence.

If we don't Round() here then we reproduce the total incidence that we are putting into the model (being driven by SpectrumIncidence[32] prior to 2002). However, the problem arises as we need to supply an integer to GetCases() as we need the number of cases that are going to be infected in the next year stratified by age and sex as an integer, we cannot use a double.

• [x] Perhaps we can separate values such that any integers are supplied to GetCases() (so if a value is >1 then we remove the integer value leaving the decimal) and the remaining non-integer values and then sample (theRng->Sample()) to determine which of the remaining values will give rise to cases? This might be a straightforward method of producing the correct number of cases.

[jackolney]

The above is quite fun and certainly fills the gap in incident cases. Prevalence is still low though...

        /* Find Incidence(a,s) */
        for(size_t j=0;j<34;j++) 
            incidence[j] = theRng->Sample( (lambda * people.at(j).size() * IRR[j]) - int(lambda * people.at(j).size() * IRR[j]) ) + int(lambda * people.at(j).size() * IRR[j]);

[jackolney]

• [x] Is the Population size the culprit? What have I changed from the previous model?

[jackolney]

IT IS IN THE POPULATION

[jackolney]

Issue fixed The culprit was a crap population distribution. I was inputting the total population size each year. Which was of course producing an enormous HIV-negative population.

const double popSize [60] = {11252466,11657477,12083159,12529800,12997438,13486241,13995996,14527242,15081677,15661480,16267906,16901167,17559844,18241424,18942464,19660018,20392312,21138372,21896890,22666720,23446439,24237056,25036941,25839132,26634659,27418077,28186224,28943647,29702246,30478597,31285050,32126351,33000524,33905011,34834606,35785718,36757498,37752304,38773277,39824734,40909194,42027891,43178141,44359872,45573945,46821246,48102684,49419194,50771734,52161292,53588880,55055540,56562340,58110380,59700787,61334722,63013375,64737971,66509767,68330055};

When I should have been putting in the Initial population size and then the number of new individuals entering the model, as below:

const double popSize[60] = {11252466,400695,419630,440241,460305,481118,502226,523708,546673,570847,594209,619182,644442,667076,682797,695625,709410,722491,731131,734908,765901,776599,784373,783075,765226,722958,718529,708110,715455,729361,750663,775364,799972,814589,845603,856064,881087,915869,946729,983062,1028592,1055553,1080443,1096579,1106563,1112580,1117645,1126043,1136366,1148139,1163788,1191180,1219216,1247912,1277283,1307346,1338116,1369610,1401846,1434841};