EID Home | Ahead of Print | Past Issues | EID Search | Contact Us | Announcements | Suggested Citation | Submit Manuscript

PDF Version | Comments | Email this article

Research

Antimicrobial Drug Use and Methicillin-resistant Staphylococcus aureus, Aberdeen, 1996–2000

Dominique L. Monnet,* Fiona M. MacKenzie,† José María López-Lozano,‡ Arielle Beyaert,§ Máximo Camacho,§ Rachel Wilson,† David Stuart,† and Ian M. Gould†
*Statens Serum Institut, Copenhagen, Denmark; †Aberdeen Royal Infirmary, Aberdeen, Scotland; ‡Hospital Vega Baja, Orihuela (Alicante), Spain; and §University of Murcia, Murcia, Spain

Suggested citation for this article: Monnet DL, MacKenzie FJ, López-Lozano JM, Bayaert A, Camacho M, Wilson R, et al. Antimicrobial drug use and methicillin-resistant Staphylococcus aureus, Aberdeen, 1996–2000. Emerg Infect Dis [serial on the Internet]. 2004 Aug. Available from: http://www.cdc.gov/ncidod/eid/vol10no8/02-0694.htm

Similar to many hospitals worldwide, Aberdeen Royal Infirmary has had an outbreak of methicillin-resistant Staphylococcus aureus (MRSA). In this setting, the outbreak is attributable to two major clones. The relationships between antimicrobial use and MRSA prevalence were analyzed by time-series analysis. From June 1997 to December 2000, dynamic, temporal relationships were found between monthly %MRSA and previous %MRSA, macrolide use, third-generation cephalosporin use, and fluoroquinolone use. This study suggests that use of antimicrobial drugs to which the MRSA outbreak strains are resistant may be an important factor in perpetuating the outbreak. Moreover, this study confirmed the ecologic effect of antimicrobial drug use (i.e., current antimicrobial use) may have an effect on resistance in future patients. Although these results may not be generalized to other hospitals, they suggest new directions for control of MRSA, which has thus far proved difficult and expensive.

Antimicrobial drug resistance occurs in hospitals worldwide. One of the most globally important microorganisms is methicillin-resistant Staphylococcus aureus (MRSA), which now causes more than 40% of all S. aureus bacteremias in the United Kingdom (1). Measures to control MRSA outbreaks have concentrated on transmission of the organism and prospective screening for carriage, in combination with general infection control measures such as patient isolation, use of barrier precautions, and environmental decontamination (2). Eradicating MRSA colonization has also been used to curb the spread of MRSA. Despite these measures, incidences of MRSA continue to rise (2,3). Guidelines for controlling MRSA in hospitals rarely include information on controlling antimicrobial use, possibly because relatively little data quantify the relationships between antimicrobial use and MRSA rates, especially in outbreak situations (4–8). To date, mathematical modeling has predicted that the effect of antimicrobial prescribing patterns in an outbreak situation is likely to be slight (9).

Epidemic MRSA type 15 (EMRSA-15) is presently the most common clone in the United Kingdom, followed by EMRSA-16, both of which are termed "super-clones" because of  their potential for spreading nationally and internationally (10). Compared to other MRSA in the United Kingdom, EMRSA-15 and EMRSA-16 are more successful at surviving, colonizing, and spreading in the hospital environment (11). Both clones are typically resistant to all β-lactams, macrolides, and fluoroquinolones (10). The northeast of Scotland has seen a rapid spread of EMRSA-16 and, to a lesser extent, of EMRSA-15 during the last 7 years after they first emerged in the area's main teaching hospital, Aberdeen Royal Infirmary.

We investigate the dynamics of the MRSA outbreak at Aberdeen Royal Infirmary and possible relationships between MRSA prevalence and antimicrobial drug use, by time-series analysis. Time-series analysis creates a mathematical model to fit a series of dynamic observations to forecast future behavior on the basis of retrospective behavior. Unlike other statistical methods that assume observed data to be independent, time-series analysis takes into account the stochastic dependence of consecutive observations or autocorrelation (12,13). This method is appropriate when data are measured repeatedly at equal intervals for long periods and when these intervals are much shorter than the study period. Time-series analysis has been applied in medical specialties such as endocrinology, cardiology, environmental medicine, and the study of chronic diseases (13). The analysis of interrupted time-series or intervention analysis is considered the strongest quasi-experimental method to ascertain the longitudinal effect of healthcare interventions (13–15). Additionally, extensions of this method, e.g., transfer function modeling and econometric dynamic modeling, can take into account external factors that may influence the target series over time and can demonstrate a temporal relationship between these external factors and the target series (13–15). Because series of antimicrobial drug use data and resistance data always show an autocorrelation, this method has been proposed by López-Lozano et al. to study the relationship between antimicrobial drug use and resistance (16).

Materials and Methods

Aberdeen Royal Infirmary is a 1,200-bed tertiary referral hospital covering a population of approximately 500,000. It comprises various medical and surgical specialties and is close to other specialized hospitals. For each month of the study period, January 1, 1996, to December 31, 2000, numbers of inpatient-days per ward were obtained from the hospital's admission department. During the study period, all S. aureus isolated were screened for susceptibility to methicillin by the comparative disc susceptibility test method on nutrient agar at 30°C with 48 h incubation (17). Methicillin resistance was confirmed by carrying out an Etest MIC. Susceptibility of the S. aureus isolates to a range of additional antimicrobial drugs was established by the comparative disc test method (17). Additionally, the Aberdeen MRSA outbreak was investigated by the Scottish MRSA Reference Laboratory, which conducted independent confirmation and genotyping. The Reference Laboratory carried out multiplex polymerase chain reaction (PCR) with primers to mecA, nuc, rRNA, 16S rRNA, (18–20) and pulsed-field gel electrophoresis (PFGE) typing of SmaI digested DNA (21).

Monthly data for all S. aureus on which antimicrobial drug susceptibility tests were carried out were exported from the clinical microbiology information system into a database. Information stored included patient identifier, hospital, ward, specimen type, and antimicrobial drug–susceptibility pattern. Because we did not systematically and uniformly search for MRSA carriers, isolates obtained from surveillance screening were excluded. Only the first S. aureus isolate from each patient within 7 days was exported from the clinical microbiology laboratory information system into an Access (Microsoft, Redmond, WA) database. Variations in the antimicrobial susceptibility pattern of S. aureus isolates from the same patient within the 7-day period were not considered. From these data, the monthly prevalence of MRSA isolates was calculated as a percentage, where the denominator was the total number of S. aureus tested for methicillin resistance.

Monthly quantities of all antimicrobial drugs delivered to each hospital ward during the study period were exported from the pharmacy information system and stored both at the individual antimicrobial drug and class level in an Access (Microsoft) database. Antimicrobial drug use was expressed as a number of defined daily doses (DDDs) per 1,000 patient days, where the DDD for each antimicrobial drug was defined by the World Health Organization (WHO) (22). As in most hospitals, data on patient exposure to antimicrobial drugs were not available at Aberdeen Royal Infirmary. For a specific antimicrobial drug class, however, the number of DDDs approximates the average number of patients exposed to an antimicrobial drug from this class each day. This measurement is the unit WHO recommends to express ecologic pressure attributable to antimicrobial drugs (23).

Time-series analysis was carried out to explore the relationships between each antimicrobial drug use series and the %MRSA series. For this purpose, linear transfer function models were built according to the identification method proposed by Pankratz (15). This analysis was completed by a graphic exploration of the series. Line plots at monthly time intervals were produced for the %MRSA and for use of each antimicrobial drug class to visualize their evolution over time and to confirm the relationships between %MRSA and antimicrobial drug use.

Once the basic characteristics (i.e., autocorrelation, seasonality, and general trend) of each of the %MRSA and antimicrobial drug use series were established, a multivariate analysis was performed to quantify the relationships between use of several antimicrobial classes and %MRSA through the use of econometric dynamic time-series modeling techniques (14,24,25). Specifically, polynomial distributed lag (PDL) modeling was used to detect and quantify lagged effects of antimicrobial drug use on %MRSA. The details of the modeling technique are presented in the Appendix. For the purposes of this study, data were analyzed with Eviews 4.0 (Quantitative Micro Software, Irvine, California, USA).

Results

	Figure 1
	Click to view enlarged image Figure 1. Evolution of the monthly number of clinical nonduplicate Staphylococcus aureus...
	Figure 2
	Click to view enlarged image Figure 2. Examples of graphic exploration of the relationship between the monthly % methicillin-resistant...
	Figure 3
	Click to view enlarged image Figure 3. Evolution of the monthly % methicillin-resistant Staphylococcus aureus (MRSA)...
	Appendix Figure
	Click to view enlarged image Appendix Figure. Examples of correlations between the monthly % methicillin-resistant Staphylococcus aureus...

From January 1996 through December 2000, the clinical microbiology laboratory isolated 9,441 nonduplicate, nonsurveillance S. aureus, including MRSA and methicillin-susceptible S. aureus (MSSA), from 6,412 hospitalized patients. Numbers ranged from 97 to 241 S. aureus isolates per month and demonstrated no seasonal patterns (Figure 1). The annual %MRSA from 1996 to 2000 were 0.6, 5.0, 14.9, 24.1, and 31.9, respectively. MRSA were rarely isolated before December 1996; after that date, a sustained increase was observed, with marked peaks of %MRSA observed in April 1998 (22%), April 1999 (30.5%), and February 2000 (38.2%) (Figure 1). Basic time-series analysis techniques and graphic exploration showed a spring seasonal variation of MRSA but no such seasonal variation for MSSA (Figure 1). From 1997 to 2000, the epidemic clones, EMRSA-16 and EMRSA-15, represented 80.0% and 15.4%, respectively, of 584 MRSA strains submitted for genotyping to the Scottish MRSA Reference Laboratory. Both clones were typically resistant to all β-lactams, macrolides, and fluoroquinolones but otherwise susceptible to other agents tested. The percentage of co-resistance to other antimicrobial drugs in all nonduplicate, nonsurveillance MRSAs (EMRSA-16, EMRSA-15, and other MRSA) isolated at Aberdeen Royal Infirmary during the outbreak is presented in Table 1. From 1996 to 2000, the annual use of systemic antibacterial agents showed little variation: 837, 953, 919, 963, and 938 DDD/1,000 patient-days, respectively. However, major variations occurred in the monthly use and seasonality of individual classes of antimicrobial drugs (Table 2).

Time-series analysis showed that %MRSA had a relationship with the use of many antimicrobial drug classes. The relationship was strongest for macrolides, fluoroquinolones, and penicillins with β-lactamase inhibitors, whereas other classes showed a significant but weaker relationship (Table 3). Graphic exploration confirmed these findings and pointed at third-generation cephalosporin use as another series to be introduced in the initial multivariate model (Figure 2). We also examined scatter plots and correlations of %MRSA with use of individual classes of antimicrobial drugs with up to 8-month delays (Appendix Figure). However, this last approach proved less useful than time-series analysis, and graphic exploration of the time series in identifying relationships and optimal delays between antimicrobial drug use and %MRSA and could be misleading. For example, scatter-plots and correlations showed an inverse correlation between MRSA and tetracycline use. However, graphic exploration showed that this correlation reflected opposite general trends rather than monthly parallel variations between these two variables (Figure 2).

A multivariate PDL model was built to relate %MRSA with use of these classes of antimicrobial drugs. The final model included previous monthly %MRSA as well as use of macrolides, third-generation cephalosporins, and fluoroquinolones as independent variables responsible for variations in %MRSA (Table 4). The greatest total effect of antimicrobial drug use on the %MRSA was found within the first two or three significant lag periods, after which the effect progressively decreased to reach nonsignificant values a few months after the end of the direct effect.

The sum of the direct and indirect effects of 10 DDD/1,000 patient-days or 30 more patients treated with a macrolide (Table 4) was an increase in %MRSA by the value 2.84 after 8 months. This change in antimicrobial drug use had more effect on the %MRSA in 1997 than in 2000. For example, in June 1997 the %MRSA was 3.6%. According to our model, an increase in macrolide use of 10 DDD/1,000 patient-days, or 30 more treated patients, made the %MRSA rise to 3.6 + 2.84 = 6.4% after 8 months or an 81% increase over June 1997. In June 2000, the %MRSA had reached 32.1%. An increase in macrolide use of 10 DDD/1,000 patient-days, or 30 more treated patients, made the %MRSA rise to 32.1 + 2.84 = 34.9% after 8 months or a 9% increase over June 2000. This observation suggests that antimicrobial drug use was a more important ecologic risk factor at the start of the outbreak than once MRSA had become endemic in the hospital. However, macrolide use kept increasing during the study period (Figure 2), which compensated for the decrease in the size of the effect of antimicrobial drug use on %MRSA. Similar effects were observed for third-generation cephalosporin and fluoroquinolone use, i.e., an increase of 10 DDD per 1,000 patient-days on a certain month or 30 more treated patients, resulted in an increase in %MRSA by 4.99 after 12 months for third-generation cephalosporins and by 4.40 after 11 months for fluoroquinolones.

The determination coefficient (R²) of the final model was 0.902, i.e., 90.2% of the variations of the monthly %MRSA from June 1997 to December 2000 were explained by the model. The model that did not take antimicrobial drug use into account (i.e., considered previous monthly %MRSA) had a lower determination coefficient (0.811) and over- or underestimated the monthly %MRSA by 7.93%. The model that took into account both previous monthly %MRSA and previous use of the three key classes of antimicrobial drugs, with a determination coefficient of 0.902, produced an average discrepancy of 2.84 percentage points with the observed %MRSA. Therefore, taking antimicrobial drug use into account helped to improve the precision in forecasting the monthly %MRSA by 64%, which is a clear indication that antimicrobial drug use has a substantial causal effect on the %MRSA.

We compared coresistance patterns of MRSA isolates from the outbreak (i.e., 1997–2000) and of MSSA from the same period (Table 1), which confirmed the consistency of the antimicrobial drug use included in the model. MRSA isolates from the outbreak period were almost always resistant to erythromycin, clindamycin, and ciprofloxacin, whereas MSSA isolates from the same period were resistant in 14.5%, 12.4%, and 35.5% of cases, respectively. Resistance of MRSA isolates to the other antimicrobial drugs tested never exceeded 11% and was lower than in MSSA isolates with the exception of mupirocin (6.1% in MRSA isolates, 1.9% in MSSA isolates).

Finally, a curve of the summed monthly use of macrolides, third-generation cephalosporins, and fluoroquinolones, which took into account their respective lags for direct effects, was constructed and plotted on the same graph as monthly %MRSA (Figure 3). This figure shows the striking parallel nature of the relationship between the lagged use of these specific antimicrobial classes and the %MRSA at Aberdeen Royal Infirmary, which confirms the findings visually.

Discussion

For the first time, a powerful statistical model provides evidence of a strong temporal relationship between antimicrobial drug use and the varying prevalence of MRSA over time during an outbreak in a single hospital. The fact that only three classes of antimicrobial drugs, namely third-generation cephalosporins, fluoroquinolones, and macrolides, showed this relationship is not surprising. In the past, exposures to cephalosporins (26,27), fluoroquinolones (27–32), and macrolides (30) have been reported as patient risk factors for MRSA infection or colonization. And cephalosporin (4,8,33,34), fluoroquinolone (5,8,33), and macrolide use (8) have been reported  as ecologic risk factors for high, or parallel variations of, MRSA prevalence or incidence. At Aberdeen Royal Infirmary, MRSA isolates were typically resistant to macrolides and fluoroquinolones (Table 1). Additionally, third-generation cephalosporins have poor activity against MRSA. At the same time, macrolides (clarithromycin and erythromycin), third-generation cephalosporins (mainly cefotaxime), and fluoroquinolones (essentially ciprofloxacin) were among the most used antimicrobial drugs at Aberdeen Royal Infirmary (Table 2), thus providing MRSA isolates with an ecologic advantage over other bacteria. Although the Aberdeen Royal Infirmary MRSA isolates were almost always resistant to clindamycin, use of lincosamides was among the lowest, which might explain why it did not appear as a risk factor in the multivariate model.

In addition to antimicrobial drug use, the final model also included the %MRSA observed 1 month before. As mentioned, we did not uniformly look for MRSA colonization. The pressure attributable to MRSA-colonized patients is a known risk factor for MRSA acquisition (8,35), which in turn affects the number of MRSA infections and the %MRSA in S. aureus from clinical samples. We therefore think that the %MRSA observed 1 month before is a surrogate for the pressure attributable to MRSA-positive patients during the past month.

The study was an ecologic and uncontrolled observational study in a single hospital. Selection bias was unlikely because data represented all hospitalized patients. Information bias was unlikely because data were not specifically collected for our study but for other purposes, i.e., routine clinical microbiologic diagnosis for S. aureus data and pharmacy accounting for antimicrobial drug use data. Confounding factors cannot be excluded but are unlikely for two reasons. First, as a result of the applied modeling strategy, the monthly variation in %MRSA not explained by the model (9.8%) was random. Therefore, the role of any possible unidentified confounding variable is thought to be minimal. Second, infection control policies, including measures such as barrier nursing, single room isolation, and eradication of carriage have consistently been applied to all MRSA patients during the study period, although a shortage of single rooms often necessitated several MRSA-positive patients being assigned to a single nurse. Staff MRSA carriers were not actively sought, but use of gloves and hand washing, as appropriate, were constantly emphasized. Active patient contact tracing was applied, when possible, but environmental cleaning relied on standard cleaning schedules rather than environmental screening and targeted interventions. This policy was in line with national guidelines (36). The relationships between antimicrobial drug use and the %MRSA were unlikely to be attributable to chance because p values in the model were low. Additionally, the cause-effect relationships in the model were validated by their temporal nature (i.e., use of macrolides, third-generation cephalosporins, and fluoroquinolones always preceded %MRSA). Additionally, for each of these antimicrobial drug classes, the effect of antimicrobial drug use on the %MRSA was directional (i.e., an increase in use resulted in increased %MRSA and a decrease in use resulted in decreased %MRSA). In contrast, variations in glycopeptide use followed variations in %MRSA with an average delay of 1 month (coefficient = 0.45).

The relative importance of antimicrobial drug use compared to cross-transmission or changes in the patient case-mix could not be assessed. In ecologic analyses with aggregated data, additional data, such as volumes of medicated soaps or alcoholic solutions used for hand hygiene, could be used as surrogates for infection control practices; however, these data were not available. As in many hospitals, patient-level data were not available, which is why we modeled aggregated microbiology and pharmacy data. Models that use patient-level data on both antimicrobial drug exposure and MRSA may reach different conclusions. For example, the risk period for a patient for acquiring MRSA then developing an infection would be limited to hospital stay, which is generally short and rarely longer than 1 month. However, our model showed that a delay of several months was sometimes necessary to observe an ecologic effect of antimicrobial drug use on the %MRSA. This result is difficult to interpret since it means that antimicrobial drug exposure of some patients on a certain month has an impact on MRSA infections in other patients several months later. Since antimicrobial drug use data are based on dispensations to the wards, antimicrobial drugs can be stocked in the wards and used over several months. However, pharmacy data showed that antimicrobial agents were dispensed several times per ward each month, making this explanation unlikely. Another explanation could be that the increase in antimicrobial drug use would contribute to increasing the size of the reservoir of MRSA carriers. First, MRSA clones would be selected in antimicrobial drug-exposed patients. Then, the size of the reservoir of MRSA carriers would gradually increase through the spread of these MRSA clones to other patients, hospital staff, and the environment. This increase would become evident in clinical samples after several months when the MRSA reservoir reached a certain size. For fluoroquinolones, this hypothesis is supported by the results of Bisognano et al. (37) and Harbarth et al. (31). These authors showed that sub-MIC levels of ciprofloxacin increase adhesion of quinolone-resistant MRSA, which could explain persistent MRSA carriage and failure of mupirocin treatment in patients who received a fluoroquinolone. Antimicrobial drug use and cross-transmission probably work together to influence the %MRSA, and if all cross-transmission were to stop after implementing a very successful control program, the relationship between fluoroquinolone use and %MRSA would most probably disappear. Further research is needed to confirm this hypothesis and, more generally, to understand why long delays are also observed for other antimicrobial drugs, e.g., third-generation cephalosporins.

At Aberdeen Royal Infirmary, antimicrobial drug prescribing is overseen by an antibiotic committee, which provides and regularly updates a joint hospital-community antibiotic policy and stewardship program (38). Antimicrobial prescribing audits are performed periodically, but changing prescribing practices to control MRSA has not been attempted. Third-generation cephalosporin prescribing was addressed previously during an outbreak of Klebsiella pneumoniae displaying extended-spectrum β-lactamase activity (39). With the implementation of the British Thoracic Society guidelines for treatment of community-acquired pneumonia (40), macrolide and third-generation cephalosporin (mainly cefotaxime) prescribing has increased, which has been paralleled by the increase in MRSA. As the Aberdeen MRSA clones are relatively susceptible, a policy of therapeutic substitution has been implemented in MRSA problem areas, starting with the replacement of cephalosporins by non–β-lactam antimicrobial drugs in surgical prophylaxis. The increase in fluoroquinolone prescribing has not been explained, but audits indicate that it is commonly used to treat serious nosocomial gram-negative infection (38).

Our study showed a quantifiable, temporal relationship between use of three classes of antimicrobial drugs (macrolides, third-generation cephalosporins, and fluoroquinolones) and the %MRSA. Because the study was performed in one hospital during an outbreak in which two predominant strains were circulating, it might not apply to other hospitals. Nevertheless, the use of antimicrobial drugs other than anti-staphylococcal penicillins and to which the MRSA outbreak strains are resistant might be a factor that would promote the outbreak. Moreover, the ecologic effect of antimicrobial drug use was confirmed (i.e., current antimicrobial drug use might have an effect on resistance in future patients). The effect of antimicrobial use on the %MRSA was greatest when the outbreak started and decreased when the %MRSA increased. Large decreases in antimicrobial drug use would have been needed to affect MRSA once it had become endemic. However, programs to control prescriptions of selected antimicrobial drug classes could represent an adjunct measure to active surveillance cultures and barrier precautions for the control of clonal outbreaks of MRSA, which has proved difficult and expensive.

Dr. Monnet is a pharmacist and microbiologist working at the National Center for Antimicrobials and Infection Control, Statens Serum Institut, Copenhagen, Denmark, as part of the Danish Integrated Antimicrobial Resistance Monitoring and Research Programme. His research interests include surveillance of antimicrobial resistance, surveillance of antimicrobial use, and the relationship between antimicrobial use and resistance.

References

1. European Antimicrobial Resistance Surveillance System. EARSS annual report 2002 [Online]. [cited 2004, Feb 24]. Available from: http://www.earss.rivm.nl/PAGINA/DOC/rep2002/annual-report-2002.pdf
2. Muto CA, Jernigan JA, Ostrowsky BE, Richet HM, Jarvis WR, Boyce JM, et al. SHEA guideline for preventing nosocomial transmission of multidrug-resistant strains of Staphylococcus aureus and Enterococcus. Infect Control Hosp Epidemiol. 2003;24:362–86.
3. Farr M, Jarvis WR. Would active surveillance cultures help control healthcare-related methicillin-resistant Staphylococcus aureus infections? Infect Control Hosp Epidemiol. 2002;23:65–8.
4. Fukatsu K, Saito H, Matsuda T, Ikeda S, Furukawa S, Muto T. Influences of type and duration of antimicrobial prophylaxis on an outbreak of methicillin-resistant Staphylococcus aureus and on the incidence of wound infection. Arch Surg. 1997;132:1320–5.
5. Manhold C, von Rolbicki U, Brase R, Timm J, von Pritzbuer E, Heimesaat M, et al. Outbreaks of Staphylococcus aureus infections during treatment of late onset pneumonia with ciprofloxacin in a prospective, randomized study. Intensive Care Med. 1998;24:1327–30.
6. Monnet DL. Methicillin-resistant Staphylococcus aureus and its relationship to antimicrobial use: possible implications for control. Infect Control Hosp Epidemiol. 1998;19:552–9.
7. Monnet DL, Frimodt-Møller N. Antimicrobial-drug use and methicillin-resistant Staphylococcus aureus. Emerg Infect Dis. 2001;7:161–3.
8. Muller AA, Mauny F, Bertin M, Cornette C, Lopez-Lozano JM, Viel JF, et al.  Relationship between spread of methicillin-resistant Staphylococcus aureus and antimicrobial use in a French university hospital. Clin Infect Dis. 2003;36:971–8.
9. Sebille V, Chevret S, Valleron A-J. Modelling the spread of resistant nosocomial pathogens in an intensive care unit. Infect Control Hosp Epidemiol. 1997;18:84–92.
10. Johnson AP, Aucken HM, Cavendish S, Ganner M, Wale MC, Warner M, et al. Dominance of EMRSA-15 and -16 among MRSA causing nosocomial bacteraemia in the UK: analysis of isolates from the European Antimicrobial Resistance Surveillance System (EARSS). J Antimicrob Chemother. 2001;48:143–4.
11. Moore PCL, Lindsay JA. Molecular characterisation of the dominant UK methicillin-resistant Staphylococcus aureus strains, EMRSA-15 and EMRSA-16. J Med Microbiol. 2002;51:516–21.
12. Box GEP, Jenkins GM. Time-series analysis: forecasting and control, revised edition. San Francisco: Holden-Day; 1976.
13. Helfenstein U. Box-Jenkins modelling in medical research. Stat Methods Med Res. 1996;5:3–22.
14. Greene WH. Econometric analysis. 3rd ed. Upper Saddle River (NJ): Prentice Hall; 1997.
15. Pankratz A. Forecasting with dynamic regression models. New York: Wiley; 1991.
16. López-Lozano JM, Monnet DL, Yagüe A, Burgos A, Gonzalo N, Campillos P, et al. Modelling and forecasting antimicrobial resistance and its relationship to antimicrobial use: a time-series analysis. Int J Antimicrob Agents. 2000;14:21–31.
17. British Society for Antimicrobial Chemotherapy Working Party. A guide to sensitivity testing. J Antimicrob Chemother. 1991;27(Suppl D):22–45.
18. Bignardi GE, Woodford N, Chapman A, Johnson AP, Speller DC. Detection of the mec-A gene and phenotypic detection of resistance in Staphylococcus aureus isolates with borderline or low-level methicillin resistance. J Antimicrob Chemother. 1996;37:53–63.
19. Brakstad OD, Aasbakk K, Maeland JA. Detection of Staphylococcus aureus by polymerase chain reaction amplification of the nuc gene. J Clin Microbiol. 1992;30:1654–60.
20. Kobayashi N, Wu H, Kojima K, Taniguchi K, Urasawa S, Uehara N, et al. Detection of mecA, femA, and femB genes in clinical strains of staphylococci using polymerase chain reaction. Epidemiol Infect. 1994;3:259–66.
21. Leonard RB, Mayer J, Sasser M, Woods ML, Mooney BR, Brinton BG, et al. Comparison of MIDI Sherlock system and pulse-field gel electrophoresis in characterizing strains of methicillin-resistant Staphylococcus aureus from a recent hospital outbreak. J Clin Microbiol. 1995;33:2723–7.
22. WHO Collaborating Centre for Drug Statistics Methodology. Anatomical Therapeutic Chemical (ATC) classification index with defined daily doses (DDDs). Oslo, Norway: The Center; 2001.
23. Capellà D. Descriptive tools and analysis. WHO Reg Publ Eur Ser. 1993;45:55–78.
24. Brown RL, Durbin J, Evans JM. Techniques for testing the constancy of regression relationships over time. J Royal Stat Soc, Series B. 1975;37:149–92.
25. Granger CWJ. Modelling economics series: readings in econometric methodology. Oxford: Oxford University Press; 1990.
26. Carmeli Y, Castro J, Eliopoulos GM, Samore MH. Clinical isolation and resistance patterns of and superinfection with 10 nosocomial pathogens after treatment with ceftriaxone versus ampicillin-sulbactam. Antimicrob Agents Chemother. 2001;45:275–9.
27. Hill DA, Herford T, Parratt D. Antibiotic usage and methicillin-resistant Staphylococcus aureus: an analysis of causality. J Antimicrob Chemother. 1998;42:676–7.
28. Campillo B, Dupeyron C, Richardet JP. Epidemiology of hospital-acquired infections in cirrhotic patients: effect of carriage of methicillin-resistant Staphylococcus aureus and influence of previous antibiotic therapy and norfloxacin prophylaxis. Epidemiol Infect. 2001;127:443–50.
29. Dziekan G, Hahn A, Thüne K, Schwarzer G, Schäfer K, Daschner FD, et al. Methicillin-resistant Staphylococcus aureus in a teaching hospital: investigation of nosocomial transmission using a matched case-control study. J Hosp Infect. 2000;46:263–70.
30. Graffunder EM, Venezia RA. Risk factors associated with nosocomial methicillin-resistant Staphylococcus aureus (MRSA) infection including previous use of antimicrobials. J Antimicrob Chemother. 2002;49:999–1005.
31. Harbarth S, Liassine N, Dharan S, Herrault P, Auckenthaler R, Pittet D. Risk factors for persistent carriage of methicillin-resistant Staphylococcus aureus. Clin Infect Dis. 2000;31:1380–5.
32. Weber SG, Gold HS, Hooper DC, Karchmer AW, Carmeli Y. Fluoroquinolones and the risk for methicillin-resistant Staphylococcus aureus in hospitalized patients. Emerg Infect Dis. 2003;9:1415–22.
33. Crowcroft NS, Ronveaux O, Monnet DL, Mertens R. Methicillin-resistant Staphylococcus aureus and antimicrobial use in Belgian hospitals. Infect Control Hosp Epidemiol. 1999;20:31–6.
34. Landman D, Chockalingam M, Quale J. Reduction in the incidence of methicillin-resistant Staphylococcus aureus and ceftazidime-resistant Klebsiella pneumoniae following changes in a hospital antibiotic formulary. Clin Infect Dis. 1999;28:1062–6.
35. Merrer J, Santoli F, Appéré-De Vecchi C, Tran B, De Jonghe B, Outin H. "Colonization pressure" and risk of acquisition of methicillin-resistant Staphylococcus aureus in a medical intensive care unit. Infect Control Hosp Epidemiol. 2000;21:718–23.
36. Advisory Group on Infection. Scottish Infection Manual: guidance on core standards for the control of infection in hospitals, health care premises and at the community interface. The Scottish Office, Department of Health; 1998.
37. Bisognano C, Vaudaux P, Rohner P, Lew DP, Hooper DC. Induction of fibronectin-binding proteins and increased adhesion of quinolone-resistant Staphylococcus aureus by subinhibitory levels of ciprofloxacin. Antimicrob Agents Chemother. 2000;44:1428–37.
38. Gould IM, Jappy B. Trends in hospital antibiotic prescribing after introduction of an antibiotic policy. J Antimicrob Chemother. 1996;38:895–904.
39. Hobson RP, MacKenzie FM, Gould IM. An outbreak of multiply-resistant Klebsiella pneumoniae in the Grampian region of Scotland. J Hosp Infect. 1996;33:249–62.
40. British Thoracic Society. British Thoracic Society guidelines for the management of community-acquired pneumonia in adults admitted to hospital. Br J Hosp Med. 1993;49:346–50.

Appendix

Polynomial Distributed Lag (PDL) model

A PDL model was built to detect and quantify the lagged effects of antimicrobial use on the % methicillin-resistant Staphylococcus aureus (MRSA). In a PDL model, the relationship between the dependent variable (resistance) and the independent variables (past resistance and antimicrobial use) should evolve smoothly over time, through the use of "polynomial lags." The optimum PDL model was arrived at by the "general-to-specific" econometric methodologic characteristics. This meant that, initially, many possible independent variables were included in the model, some of which were ultimately found to be irrelevant. Additionally, for all the independent variables, lags of up to 8 months were initially included to identify direct effects. The initial dynamic regression model with PDLs considering %MRSA series as the dependent variable and several antimicrobial drug use series as explanatory series was the following:

with PDL restrictions on the coefficients of antimicrobial use and where MAC means macrolide use, 3GC third-generation cephalosporin use, FQ fluoroquinolone use and PIB use of penicillins with β-lactamase inhibitors. The model was initially estimated on the full study period, i.e., January 1996–December 2000, using a degree q_j of the polynomial equal to 3. The estimated model was compatible with normal white noise errors (absence of autocorrelation and absence of heteroskedasticity), and no signs of nonmodeled nonlinearities were seen.

This initial model was then simplified to eliminate irrelevant antimicrobial drug uses and unnecessary lags. In the first steps of the simplification, all antimicrobial drugs were kept in the model, and the simplification took the form of reducing the order of the polynomial and eliminating unnecessary lags. Along this process, use of penicillins with β-lactamase inhibitors did not appear to play a significant role and was eliminated from the model. We also tried to introduce use of each of the other antimicrobial drug classes that showed a relationship in Table 3; however, none appeared to play an important role, and they were not included in the model. Further simplification of the distributed lags of macrolide use, third-generation cephalosporin use, and fluoroquinolone use of the %MRSA itself led to a model in which, through CUSUM and CUSUMSQ statistics, a structural change was detected around the middle of 1997. Application of the Chow test located the change in June 1997. The %MRSA was virtually zero in 1996 and started to increase at the beginning of 1997, which was consistent with the fact that the MRSA epidemic strain, resistant to macrolides and fluoroquinolones, only became predominant in 1997. In 1996, 56% and 50% of MRSA isolates were resistant to erythromycin and ciprofloxacin, respectively, whereas these percentages suddenly rose to 92% and 89%, respectively, in 1997. Data before June 1997 were considered as not being part of the outbreak and were therefore not included in the final model. The validity of the simplified, final model from June 1997 onwards was checked by a battery of specification and diagnostic tests to verify the absence of autocorrelation of residuals, absence of heteroskedasticity, normality of residuals, absence of nonmodeled nonlinearities and absence of structural change.

The basic measure of forecasting quality, Root Mean Squared Error of Forecast (RMSEF) was also computed, which provided an average measurement of the amount by which the model over- or underestimated the %MRSA. RMSEF was calculated for a model without antimicrobial drug use (based on past %MRSA only) and compared with that of the final model, which included antimicrobial drug use.

EID Home | Top of Page | Ahead-of-Print | Past Issues | Suggested Citation | EID Search | Contact Us | Accessibility | Privacy Policy Notice | CDC Home | CDC Search | Health Topics A-Z

This page posted July 6, 2004
This page last reviewed July 22, 2004

Emerging Infectious Diseases Journal
National Center for Infectious Diseases
Centers for Disease Control and Prevention