[Gnumed-devel] Interactions and stuff: proposed rules database.

[Ian Haywood]

I wonder if anyone has seen the article in last week's OzDoc?
[For the Germans: a short article complaining about how prescribing
software pops up too many warnings, yet don't pick up all interaction. 
It blames 'class-based interactions'
and also laments lack of support for the triple interaction 
diuretic+ACEI+NSAID]

Hilmar has also rightly pointed out the need to specify by 
route, dose, etc.

The problem with these features is that the drug db would become
unreasonably complex.

Furthermore, there are other rules we would add to GNUMed, such as reminders 
for Pap
smears, etc, requiring their own SQL structures. It would be nice if these 
clinical
knowledge items were subject to the same peer-review we will have for the drug 
database.

Proposal:

A 'rules database':

A table of two text fields: a doctor's warning/suggestion
(a free (localised) string), and a boolean expression 
identifying the circumstances of the warning. I propose
a COBOL-ish mini-language to specify this.
It would have predicates such as 

PRESCRIBED [drug] 
DIAGNOSED [disease]
DOSE < | >  [amount] [unit]  } on previous PRESCRIBED clause
ROUTE [route]                }
TESTED [investigation]
KEYWORD [keyword in clinical history]
AGE < | > [time]
FEMALE
MALE
PREGNANT
LACTATING

Operators:
AND, OR, NOT, SINCE [time]

Examples

NOT TESTED pap-smear SINCE 2 years

PRESCRIBED penicillin AND DOSE > 1 G AND ROUTE IV AND PRESCRIBED warfarin

FEMALE AND AGE > 14 AND AGE < 60 AND PRESCRIBED isotrentanoin


The rule database is queried at key points in the client (prescription, 
entering clinical note, 
new diagnosis, also as a separate command to print reminder letters. 
The queries can be done by a SQL stored procedure in C, which can pre-select 
queries 
(if a penicillin is prescribed, only expressions containing "PRESCRIBED 
penicillin" need to be
checked)