PROBLEMS WHEN ATTEMPTING TO DEMONSTRATE A RELATIONSHIP BETWEEN ANTIBIOTIC CONSUMPTION AND BACTERIAL RESISTANCE

1. PROBLEMS WHEN ATTEMPTING TO DEMONSTRATE A RELATIONSHIP BETWEEN ANTIBIOTIC CONSUMPTION AND BACTERIAL RESISTANCE

Data must be of the same type for both antibiotic consumption and bacterial resistance if one wants to study the relationship between these two parameters. These can be collected at the patient or at the collective level. Patient-level data allow the study of the effect of individual patient exposure to antibiotics on emergence and selection of bacterial resistance in this patient. However, these data are rarely available and do not take into account the possible conse- quences on resistance in other patients. Aggregated data, that is, data at the level of a hospital, a ward, or a primary care region, cannot take into account

Antibiotic Policies: Theory and Practice. Edited by Gould and van der Meer Kluwer Academic / Plenum Publishers, New York, 2005

448 José-María López-Lozano et al. misuse of antibiotics in individual patients and only represent the ecological

pressure due to antibiotics; however, they often are the only available data in most hospitals. The type of aggregated data will determine the type of evi- dence in the demonstration of a relationship. If data are available for a specific period and for a large number of similar and independent settings, it will be possible to demonstrate a consistent association and/or a dose–effect relation- ship between antibiotic consumption and bacterial resistance (McGowan, 1987). Common problems with this approach are small sample size and possi- ble selection bias. Multicenter studies, thus increasing statistical power and minimizing selection bias, have been implemented to circumvent these prob- lems. However, even these multicenter studies often lack adequate sample size, include observations that are not independent, for example, hospitals or wards that exchange patients, cannot take into account the necessary delay between antibiotic use and bacterial resistance, and fail to deal with consumption of other antibiotic classes and other confounding factors. To illustrate this point, the data presented in Figure 1 suggest a relationship between the percentage of ceftazidime-resistant Pseudomonas aeruginosa and consumption of third- generation cephalosporins. However, this relationship clearly does not apply to two out of eight (25%) hospitals, where resistance is possibly explained by other factors such as consumption of other antibiotics, infection control practices, detection of resistance, etc. (McGowan, 1994; Monnet, 2000b).

When longitudinal data on antibiotic consumption and on bacterial resis- tance are available for a long period of time, it is possible to study concomitant variations, that is, changes in antibiotic consumption followed by changes in resistance in the same direction (McGowan, 1987). These variations are the most convincing proofs of causality when using aggregated data since they

Figure 1. Percent ceftazidime-resistant/intermediate P. aeruginosa and third-generation cephalosporin use at eight US hospitals, CDC/NNIS Project ICARE Phase 1, 1994. Adapted from Monnet et al. (1998).

Applications of Time-series Analysis 449 take into account the time sequence between the suspected cause, that is,

antibiotic consumption, and the observed effect, that is, bacterial resistance. Such concomitant variations have been reported from various single hospitals or from single countries. However, when pooled data, covering one or several years, were used to analyze temporal associations observed between two time periods, these studies were not able to measure the delay necessary to observe an effect of antimicrobial use on resistance. Additionally, because data generally are available on a yearly basis, empirical attempts to take this delay into account consider a 1-year delay. For example, Goossens et al. (1986) found a correlation between the percentage of gentamicin-resistant Gram- negative bacilli and gentamicin use during the previous year. Interestingly, no correlation was observed when using gentamicin during the same year.

Our experience at Hospital Vega Baja (López-Lozano et al., 2000b) shows that there is often a 1-year delay between a variation in antibiotic use and a consecutive variation in resistance, but not always. As shown in the example presented in Figure 2, we generally observed such variations of the percentage of ceftazidime-resistant Gram-negative bacilli following variations in cef- tazidime use, except in 1996 when ceftazidime resistance increased following

a decrease in ceftazidime use in 1995. However, if we graphically represent the monthly percentage of ceftazidime- resistant Gram-negative bacilli and the monthly consumption of ceftazidime, we can observe variations at time periods shorter than 1 year (Figure 3). As shown below, there is a significant relationship between these monthly data.

Additionally, by simply smoothing the monthly series using a 5-month centered moving average transformation, the relationship becomes clear with ceftazidime consumption preceding ceftazidime resistance (Figure 4).

Figure 2. Yearly hospital ceftazidime use and percent ceftazidime-resistant/intermediate Gram-negative bacilli, Hospital Vega Baja, Orihuela, Spain, 1991–8. Adapted from Monnet et al. (2001). 䊏, ceftazidime use (DDD/1,000 patient-days); 䊐, ceftazidime-resistant/ intermediate Gram-negative bacilli (%).

450 José-María López-Lozano et al.

Figure 3. Monthly hospital ceftazidime use and percent ceftazidime-resistant/intermediate Gram-negative bacilli, Hospital Vega Baja, Orihuela, Spain, 1991–8. Adapted from López- Lozano et al. (2000).

, ceftazidime use (DDD/1,000 patient-days); ___, ceftazidime- resistant/intermediate gram-negative bacilli (%).

Figure 4. Five-period centered moving average of monthly hospital ceftazidime use and per- cent ceftazidime-resistant/intermediate Gram-negative bacilli, Hospital Vega Baja, Orihuela, Spain, 1991–8. Adapted from López-Lozano et al. (2000).

, ceftazidime use (DDD/1,000 patient-days); ___, ceftazidime-resistant/intermediate Gram-negative bacilli (%).