Not sure this is something R could do but it feels like it should be.

Some colleagues nationally have developed a system which means they can pick the optimal sets of doses for a drug.  The system could apply to a number of drugs.  But the actual doses might vary.  To try and explain this in terms that the average Joe on the street might understand if you have some amoxicillin antibiotic for a chest infection the normal dose for an adult is 250 to 500mg increased to maybe 1000mg in severe cases.

For a child it is dosed from a liquid and people usually go from 62.5mg, 125mg to 250mg although you could measure any volume you wanted.

What this new method has developed is a means to pick the "right" standard doses so what above is 62.5, 125, 250, 500, 1000.  However the method they've used is really engineered about ensure the jump between doses is correct - you'll notice that the list above is a doubling up method.

But you can also have a doubling up method that went 50, 100, 200, 400, 800, 1600  and pretty much as many as you can think of depending on your starting point and there is no scientific means to pick that starting point.  So colleagues have developed their rather more complex equivalent of the doubling method to determine the doses they need but they need to know if they should start at 40, 50, 62.5 or some other number.

Once they have the starting number they can calculate all the other doses.  I realise R can do that, and I realise using a loop of possible starting numbers it can build all those options.

Each patient then has a theoretical dose they should get lets say that's 10mg/kg and you might treat patients from 5 to 120kg.  They are then looking to calculate the variance for each dose range so if we take the 50, 100, 200, 400 model and said you'd give 50mg to anyone needing 0?? to 75mg 100mg to anyone needing 76 - 150mg etc... from there they are taking that range and saying that's a 31% overdose to a 33% underdose.  Then they want to find if there is a starting number which minimises the extent of under and overdosing...

Anyone know of an existing stats function in R that can easily do that and almost then report from some inputs a single number that is the "best fit"?

Calum

Some colleagues nationally have developed a system which means they can pick the optimal sets of doses for a drug.  The system could apply to a number of drugs.  But the actual doses might vary.  To try and explain this in terms that the average Joe on the street might understand if you have some amoxicillin antibiotic for a chest infection the normal dose for an adult is 250 to 500mg increased to maybe 1000mg in severe cases.

For a child it is dosed from a liquid and people usually go from 62.5mg, 125mg to 250mg although you could measure any volume you wanted.

What this new method has developed is a means to pick the "right" standard doses so what above is 62.5, 125, 250, 500, 1000.  However the method they've used is really engineered about ensure the jump between doses is correct - you'll notice that the list above is a doubling up method.

But you can also have a doubling up method that went 50, 100, 200, 400, 800, 1600  and pretty much as many as you can think of depending on your starting point and there is no scientific means to pick that starting point.  So colleagues have developed their rather more complex equivalent of the doubling method to determine the doses they need but they need to know if they should start at 40, 50, 62.5 or some other number.

Once they have the starting number they can calculate all the other doses.  I realise R can do that, and I realise using a loop of possible starting numbers it can build all those options.

Each patient then has a theoretical dose they should get lets say that's 10mg/kg and you might treat patients from 5 to 120kg.  They are then looking to calculate the variance for each dose range so if we take the 50, 100, 200, 400 model and said you'd give 50mg to anyone needing 0?? to 75mg 100mg to anyone needing 76 - 150mg etc... from there they are taking that range and saying that's a 31% overdose to a 33% underdose.  Then they want to find if there is a starting number which minimises the extent of under and overdosing...

Anyone know of an existing stats function in R that can easily do that and almost then report from some inputs a single number that is the "best fit"?

Calum

Calum

Hi Calum,
I can only answer from the perspective of someone who calculated
doses of alcohol for experimental subjects many years ago. It was not
possible to apply a linear function across the range due to a number
of factors. One is that BAC, which was the target value, is dependent
upon the proportion of the weight that represents the water
compartment of the body. This varies with both weight (heavier people
typically have a higher proportion of fat) and sex (women also tend 
to
have slightly more fat). The real monkey wrench in the works was
absorption rate, which often made nonsense of my calculations. This
may not be as important in therapeutic drugs, for we were aiming at a
specified BAC at a certain time after dosing rather than an average
level.

However, I suspect that many therapeutic drugs have a different
dose by weight for children (we weren't dosing children) and choosing
a starting point at the bottom of the range would almost certainly
introduce a systematic error. My intuition would be to anchor the
dosage rate in the middle of the scale and then extrapolate in both
directions (adults only, of course).

Calum

Hi Calum,
I can only answer from the perspective of someone who calculated
doses of alcohol for experimental subjects many years ago. It was not
possible to apply a linear function across the range due to a number
of factors. One is that BAC, which was the target value, is dependent
upon the proportion of the weight that represents the water
compartment of the body. This varies with both weight (heavier people
typically have a higher proportion of fat) and sex (women also tend to
have slightly more fat). The real monkey wrench in the works was
absorption rate, which often made nonsense of my calculations. This
may not be as important in therapeutic drugs, for we were aiming at a
specified BAC at a certain time after dosing rather than an average
level.

All those things affect therapeutic dosing.

I may have oversimplified what we are trying to achieve to avoid getting bogged down in the detail of what we are trying to achieve and provide something people might be able to relate to.

However, we can assume that they are already sorted out, so we know the theoretically know what the 'correct' dose is for a patient.  The hard bit is unless you want to give everyone liquid so you can measure any dose possible you have to have a dose that is a multiple of something (Amoxicillin doses in adults are multiples of 250 because thats the size of the capsule).

What we are trying to do is determine the most appropriate number to make the capsules.  (Our dosing is more complex but lets stick to something simple.  I can safely assure you that vritually no-one actually needs 250 or 500mg as a dose of amoxicillin... ...thats just a dose to get them into a therapeutic window, and I'm 99% certain 250 and 500 are used coz they are round numbers.  if 337.5 more reliably got everyone in the window without kicking anyone out the window that'd be a better dose to use!  So... what I'm looking to do is model the 'theoretical dose required' (which we know) and the dose delivered using several starting points to get the 'best fit'.  We know they need to be within 7% of each other, but if one starting point can get 85% of doses within 5% we think that might be better than one that only gets 50% within 5%.

However, I suspect that many therapeutic drugs have a different
dose by weight for children (we weren't dosing children) and choosing
a starting point at the bottom of the range would almost certainly
introduce a systematic error. My intuition would be to anchor the
dosage rate in the middle of the scale and then extrapolate in both
directions (adults only, of course).

We are actually using a starting point that may be middle and going up and down if need be.

I think what we may want to do is run a loop through each weight (in 1kg increments) and calculate their theoretical dose, and the dose for each possible starting point (there are certain contraints on that already so there may only be 20 possible start points), then we calculate the % variance for each dose to theoretical dose and calculate the Area Under & Above (some will be negative) the curve and the one that has the lowest AUC is then the one that most "precisely" will dose the patient??

...
What we are trying to do is determine the most appropriate number to
make the capsules. (Our dosing is more complex but lets stick to
something simple. I can safely assure you that vritually no-one actually
needs 250 or 500mg as a dose of amoxicillin... ...thats just a dose to
get them into a therapeutic window, and I'm 99% certain 250 and 500 are
used coz they are round numbers. if 337.5 more reliably got everyone in
the window without kicking anyone out the window that'd be a better dose
to use! So... what I'm looking to do is model the 'theoretical dose
required' (which we know) and the dose delivered using several starting
points to get the 'best fit'. We know they need to be within 7% of each
other, but if one starting point can get 85% of doses within 5% we think
that might be better than one that only gets 50% within 5%.