To predict the short-term evolution of resistance

5.2. To predict the short-term evolution of resistance

When a doctor is confronted with symptoms of a possible infection in

a patient several issues arise: (1) if it really is an infection, (2) is possible to expect a satisfactory spontaneous progress, (3) if infection is due to a bacteria,

a fungus, a virus, or another microorganism, and (4) in the event a bacterial origin is suspected and once the treatment is decided, the doctor must choose among the available antibiotics, the one that maximizes the possibilities of therapeutic success. Apart from clinical or pharmacologic considerations, the most influential factor in his decision is the possibility that the unknown microorganism responsible for the infection be resistant to the antibiotic he decides to use, (5) a basic factor is to determine the infecting microorganism. The problem that arises is that the microbiological analysis takes a while (one, two, three days) to show information about the germ that infects the patient, but meanwhile is often necessary to start the therapy, therefore an important factor is to predict the causal microorganism. In that sense, to know the usual

456 José-María López-Lozano et al. local flora (hospital, community, etc.) will help us. (6) On the other hand, once

it is established as to which is the microorganism that most probably caused the infection, we must choose among the available antibiotics. The knowledge about the spectrum of bacterial resistance in the setting in question (hospital, community, etc.) will help us in the choice of an antibiotic with minimal pos- sibilities of the causal germ to be resistant. The doctor usually support his deci- sion on guidelines made in other places, which recommend an antibiotic to use against this or that microorganism. Obviously, these recommendations should not be necessarily applied in our environment given that, as has been explained above, the stochastic nature of the phenomenon makes certain microorgan- isms show resistance phenotypic patterns completely different in different environments.

Time-series analysis let us make predictions based on previous values and related factors in terms of causality that we know, particularly antibiotic use. These predictions can be interesting in terms of practical application as an aid to their empiric antimicrobial treatment. Indeed, if we were able to decide every time as to what is the probability of a patient to have an infection due to this or that microorganism and if we were also able to estimate the probability of every microorganism to be sensitive or not to every available antibiotic, we could make recommendations about the empiric therapy, relevant to our local environment, that would minimize the probability of error in the therapeutic choice.

ViResiST (Spanish acronym for ‘Resistance Surveillance using Time- series Analysis Techniques’, www.viresist.org) is a project that focuses on antibiotic consumption and bacterial resistance in a short-term temporal dimension and at the local level (López-Lozano et al., 2002b). It uses the following:

1. Microbiology data: Monthly hospital and community antibiotic susceptibil- ity data for several years (usually from 1992) are exported into the ViResiST database. Duplicate and surveillance, that is, screening, isolates were excluded.

2. Pharmacy data: For the same period, monthly quantities of antimicrobials prescribed in the community and hospital are also exported into the ViResiST database. Data on use of individual antibiotics and antibiotic classes are stored as a number of DDD per 1,000 inhabitant-days for the community and per 1,000 patient-days for the hospital (ATC, 2002).

A Windows based interface allows an easy examination of the data as well as the selection and exportation of interesting time series (of monthly percent- ages of resistance and/or of monthly antibiotic consumption) in order to analyze them using time-series analysis techniques.

Applications of Time-series Analysis 457 The first usefulness of the application program is a look up table with the

results of the predictions of the ARIMA models fitted on each series of resis- tance. This table gathers the expected resistance for the current quarter as per- centages of resistant strains of each microorganism to each antibiotic. In order to calculate these predictions we fit different ARIMA models using a semiau- tomatic method based on the software SCA (www.scausa.com). The applica- tion program also calculates the frequency of each microorganism in each type of sample, hospital department, and primary healthcare facility.

This information can be used by the clinician to improve his empiric ther- apy when he suspects that the patient has an infection caused by a certain microorganism and must decide among various antibiotics. Once the microor- ganism is known, the tables give the possibility of choosing the antibiotic with the lowest expected resistance. The expected resistance percentage and the fre- quency of microorganism in each type of sample can be used for the elabora- tion of empiric antimicrobial therapy guidelines based on local ecology at a certain hospital or health center.