To evaluate interventions to control antibiotic resistance

5.3. To evaluate interventions to control antibiotic resistance

Example. Evaluation of a hospital-wide policy restricting third generation cephalosporin use to reduce Clostridium difficile-associated diarrhea: a time- series analysis (Thomas et al., 2003)

The occurrence of Clostridium difficile-associated diarrhea (CDAD) has been observed closely at Sir Charles Gairdner Hospital (SCGH), a 560-bed urban teaching hospital in West Australia, since the early 1980s. Between 1983 and 1992, the incidence of CDAD increased from 23 to 50 cases per 100,000 patient-days at a time when consumption of third generation cephalosporin (3GC) antibiotics also increased (Riley et al., 1994). Following the restriction of 3GC at the hospital in 1998, a significant decline in incidence of CDAD was observed (Thomas, 2002). The aim of this study was to describe the relationship between CDAD and 3GC consumption from 1993 to 2000, using time-series methodology, in order to evaluate the impact of the antibiotic policy change.

The study was conducted in SCGH from 1993–2000. Hospitalized patients that had a positive laboratory test, either direct faecal cytotoxin detection, or culture of C. difficile, were identified from the microbiology laboratory data- base from January 1993 to 2000. A period of 14 days or less between positive laboratory tests for C. difficile was considered to be part of the same episode.

Monthly use of 3GCs for the same period was obtained from the hospital’s pharmacy as gram amounts dispensed and converted to the number of DDD per month (ATC, 2002).

458 José-María López-Lozano et al. Polynomial distributed lag (PDL) modeling was used to evaluate the rela-

tionship between 3GC use and the incidence of CDAD episodes in SCGH. PDL models for the monthly incidence rate of CDAD episodes, expressed as the number of C. difficile positive episodes/1,000 patient-days (the dependent series), and the monthly rate of 3GC use, expressed as DDDs/1,000 patient- days (the independent series), were constructed using Eviews 4.0 (Quantitative Micro Software, Irvine, California, USA). Diagnostic checking of the models consisted of ensuring all significant autocorrelation patterns were accounted for. The cumulative sum (CUSUM) and cumulative sum of squares (CUSUMQ) plots of the residuals were examined to locate possible structural changes in the model that were confirmed by applying Chow tests. The good-

ness of fit of the model was estimated using the determination coefficient (R 2 ). Monthly incidence of CDAD and use of 3GC in SCGH, from January 1993 to December 2000 are illustrated in Figure 6. 3GC consumption fell from

28.95 DDD/1,000 patient-days (95% CI 28.63–29.26) prior to October 1998 to

3.29 DDD/1,000 patient-days (95% CI 3.12–3.46) after the policy was intro- duced. The average incidence of CDAD during the pre-intervention period was

0.61 episodes/1,000 patient-days (95% CI 0.56–0.65). During the post-inter- vention period the average incidence was 0.28 episodes/1000 patient days (95% CI 0.23–0.33).

The output for the final model of the relationship between 3GC antibiotics and CDAD is presented in Table 2. The incidence of CDAD observed in one month was related to (1) the incidence of CDAD four months previously and (2) the consumption of 3GC antibiotics during the same month. From the

40 DDD/1,000 patient-days Episodes/1,000 patient-days 0.2

Date(month-year)

CDAD

3GC

Figure 6. Five-period centered moving average of monthly incidence of C. difficile-associ- ated diarrhoea episodes and hospital third-generation cephalosporin use, Sir Charles Gairdner Hospital, Perth, Australia, 1993–2000. Adapted from Thomas et al. (2003).

Applications of Time-series Analysis 459

Table 2. Polynomial distributed lag model for monthly incidence of Clostridium difficile- associated diarrhoea (CDAD) episodes, Sir Charles Gairdner Hospital, Perth, Australia,

March 1994–December 2000 (R 2 ⫽ 0.62). Adapted from Thomas et al. (2003) Independent variable

Coefficient T-statistic P-value Past incidence of CDAD

Lag (months)

4 0.30 3.92 0.0002 episodes Third-generation cephalosporin

0 0.013 7.95 0.0000 use Structural change (09/1997)

0 0.22 6.16 0.0000 Interaction between structural

0 ⫺ 0.018 ⫺ 4.78 0.0000 change (08/1998) and third- generation cephalosporin use

coefficients in Table 2, if 3GC use increased by 1DDD/1000 patient-days in a particular month, an increase in 0.013 CDAD episodes/1000 patient-days could be expected in that same month.

Two structural changes were identified in the model. First, a shift in CDAD took place in September 1997 and, second, a substantial change in the influence of 3GC antibiotics occurred in August 1998. Prior to August 1998, 3GC use had

a positive relationship with CDAD; after August 1998 this effect was canceled. That is, when consumption of 3GC fell, the impact on CDAD was so great that

a relationship between 3GC and CDAD was no longer statistically detectable. Time-series analysis was used to describe the temporal relationship between 3GC use and the incidence of CDAD in a public teaching hospital, in order to evaluate the impact of a restriction in the use of ceftriaxone at the hospital.

Using PDL modelling, 3GC consumption in one month was related to CDAD in the same month. Past incidence of CDAD was also related to current CDAD, however, the effect of a change in the incidence in 1 month was not produced until 4 months later. Explanations for this are not clear, however,

a 4-month lag such as this relates to data from the preceding quarter and there- fore may indicate some seasonal variation. For example, variation in hospital contamination, or the use of other antibiotics for community-acquired pneu- monia during winter months.

The ceftriaxone policy implementation manifested as a structural change in the relationship between 3GC consumption and CDAD incidence from August 1998, 2 months prior to the policy introduction. An audit of 3GC was under- taken in the study hospital during July 1998 that may have influenced subse- quent use of 3GC before the policy was introduced. During the period of the study, ceftriaxone was the only 3GC antibiotic in use from early 1997 after ceftazidime and cefotaxime were discontinued. However this change did not manifest as a structural change in the PDL model. An unexplained shift in CDAD was detected in September 1997 that may be attributed to unknown

460 José-María López-Lozano et al. factors such as changes in consumption of other antimicrobials or infection

control practices. The results from this study indicate that in settings where 3GC consump- tion is high, reduction in the use of these antibiotics, -via policies aimed at pre- scription restriction, effectively reduces the occurrence of C. difficile infection in hospitals. Although others have reported similar observations for 3GCs (Ludlam et al., 1999) this is the first study to evaluate such a policy by quanti- fying the relationship of antibiotics with CDAD using time-series models.