Impact of universal screening on MRSA bacteremias in a single acute NHS organisation (2006–12): interrupted time-series analysis

Background In November 2004, a national target was set for the English hospital trusts to reduce the Meticillin-Resistant Staphylococcus aureus (MRSA) bacteremia rate by 60% by April 2008 against the number during 2003/04 (baseline year). In our organisation the number of MRSA bacteremias had risen since 2002 and peaked at 75 in 2005/06. A target was set to reduce the number and series of specific and non- specific interventions was introduced including universal MRSA screening. This study analyzes the impact of universal MRSA screening using a quasi-experimental design using routinely gathered data. Methods This study used data gathered routinely for clinical governance, quality control, financial management and outbreak monitoring purposes. Interrupted Time Series (ITS) analysis of 15 pre- and 19 post- universal MRSA screening (and decolonisation) quarterly numbers of bacteremias was carried out where Meticillin-Sensitive Staphylococcus aureus (MSSA) numbers served as non-equivalent dependent variable (control). Results An immediate sharp fall in MRSA bacteremias was observed following the universal MRSA screening (and decolonisation) commenced in Q2, 2007. The number dropped sharply from 23 (Q2, 2007) to 10 (Q3, 2007) for all MRSA bacteremias, and, from 15 (Q2, 2007) to 6 (Q3, 2007) for bacteremias ≥48 hours of hospitalization. The declining trend continued reaching zero in Q2, 2009 and Q4, 2010 for those with ≥48 hours of hospitalization and all bacteremias, respectively. ITS analysis revealed significant impact of universal MRSA screening on all MRSA bacteremias (β2 -0.554, p 0.000) and those with ≥48 of hospitalization (β2 -0.577, p 0.001). Impact estimation predicted 17 and 13 bacteremias for all and those with ≥48 hours hospitalization, respectively in the 19th quarter post-intervention, if the intervention did not occur. The number of MRSA isolates from non-blood culture systemic sources as percentage of admissions also dropped significantly from 3.32% in Q2, 2007 to 1.51% in Q3, 2007 (β2 -0.506, p 0.000) which is still running low at 0.33% at the end of Q1, 2012. On the other hand, there was no statistically significant impact of universal screening on MSSA bacteremias. Conclusions We conclude that of all interventions, the universal MRSA screening (and decolonisation) is the most effective intervention associated with significant and sharp drop in MRSA burden.

Results: An immediate sharp fall in MRSA bacteremias was observed following the universal MRSA screening (and decolonisation) commenced in Q2, 2007. The number dropped sharply from 23 (Q2, 2007) to 10 (Q3, 2007) for all MRSA bacteremias, and, from 15 (Q2, 2007) to 6 (Q3, 2007) for bacteremias ≥48 hours of hospitalization. The declining trend continued reaching zero in Q2, 2009 and Q4, 2010 for those with ≥48 hours of hospitalization and all bacteremias, respectively. ITS analysis revealed significant impact of universal MRSA screening on all MRSA bacteremias (β 2 -0.554, p 0.000) and those with ≥48 of hospitalization (β 2 -0.577, p 0.001). Impact estimation predicted 17 and 13 bacteremias for all and those with ≥48 hours hospitalization, respectively in the 19th quarter post-intervention, if the intervention did not occur. The number of MRSA isolates from non-blood culture systemic sources as percentage of admissions also dropped significantly from 3.32% in Q2, 2007 to 1.51% in Q3, 2007 (β 2 -0.506, p 0.000) which is still running low at 0.33% at the end of Q1, 2012. On the other hand, there was no statistically significant impact of universal screening on MSSA bacteremias.

Conclusions:
We conclude that of all interventions, the universal MRSA screening (and decolonisation) is the most effective intervention associated with significant and sharp drop in MRSA burden. The 'Clean your hands' campaign was already ongoing since September 2004. This study reviews the impact of these interventions to reduce the burden of MRSA bacteremias, specifically, the impact of universal MRSA screening using a quasi-experimental (Interrupted Time Series -ITS) design using routinely gathered data in the context of a comprehensive infection control program.

Methods
The setting is an acute NHS Trust in the north east of England serving a mixed urban/rural predominantly elderly population of approximately 500,000. The Trust has three acute district general and seven community hospitals with 760 acute medical/elderly care beds, and 183 acute surgical beds including orthopedics. The average number of beds per ward is 21 with 82.9% bed occupancy. There is an active infection control team (ICT) with four point five whole time equivalent microbiologists, seven infection control nurses and input from other specialties including pharmacy. The study used no patient identifiable information and is a noninterventional observational study and as such did not require ethical approval.

Data collection
This study used data gathered routinely for clinical governance, quality control, financial management and outbreak monitoring purposes. The numbers of monthly bacteremias were routinely recorded for governance purposes. The review of surveillance (Sept 2004 to March 2006) data showed the sources of MRSA bacteremias to be unknown in an unusually large proportion (30%) of cases, and in the rest, three major sources were identified viz., indwelling intra vascular devices, chest infections and soft tissue infections (Figure 1). Following the introduction of Root Cause Analysis (RCA) in May 2006 a number of interventions were made in quick succession as part of the MRSA improvement programme; implementation of these interventions and compliance levels were routinely monitored by audits (Table 1) The MRSA positive patients were isolated according to the infection control policy and commenced on decolonisation treatment as soon they were identified irrespective of whether facilities were available to isolate. Octenisan W (or 2% Triclosan) body wash once daily and Mupirocin 2% Nasal ointment three times a day to the inside of each nostril for 5 days (and hair washing with Octenisan W on days 2 and 4 of the treatment) were 19    The blood culture policy was re-issued which required that indication was recorded in the medical notes and blood culture was authorised by a consultant/senior doctor. A training video was made available on the Trust intranet to demonstrate how to take cultures using aseptic technique. This was also shown at induction for new junior doctors. In June, the number of blood cultures taken fell from the monthly average of 1252 to 778, a 38% reduction, of which 16% were positive compared to 14.5% in the previous five months. The proportion that was skin organisms (e.g., coagulase negative staphylococci, diptheroids and propionibacterium) marginally reduced to 24% from 29%. In general, up to two complete courses of decolonisation were given during each inpatient stay. Any patient who was still positive for MRSA after a second decolonisation treatment received daily washes using octenisan for the duration of their in-patient stay.

Analysis
For the interrupted time-series (ITS) analysis, we used segmented linear regression, which divides a time series into pre-and post-intervention segments. The universal screening was introduced in Q2, 2007, therefore, Q3, 2007 was chosen as the intersection between the preand post-intervention segments. Interventions that may be short-lived may erroneously report maximal effects if short time series are analyzed; our time series included 19 pre-and 15 post-intervention quarterly data points.
A linear regression model has two parameters: the level and slope. Therefore the difference between the two segments can be quantified by testing the change in these two parameters (Equation 1). A change in level between the pre-and post-intervention segments indicates a step-change, and a change in slope indicates a change in trend.
Y t is the MRSA rate at month t; β 0 estimates the baseline MRSA quarterly rate; β 1 estimates the baseline (pre-intervention) linear trend where T is a continuous variable indicating the time in quarter interval at time t from the start of the study period; β 2 estimates the level change between pre-and post-   intervention, where D = 0 before the intervention, and D = 1 after the intervention; β 3 estimates the mean quarterly trend in MRSA postintervention, compared to the baseline trend, where P is a continuous variable indicating the number of quarters after the start of the intervention at time t and is coded as zero before the intervention.
Three points of interest were calculated: (1) preintervention trend (slope); (2) post-intervention change of the baseline level representing an immediate effect; and (3) post-intervention trend (slope) representing a sustained effect of the intervention. A p-value of 0.05 was regarded as statistically significant. If coefficient β 1 was significantly different from 0, the pre-intervention trend was deemed statistically significant. Similarly, if β 2 and/or β 3 were significantly different from 0, it was assumed that there had been a significant postintervention change of baseline level and/or trend. We included Meticillin-Sensitive Staphylococcus aureus (MSSA) numbers for the same period as nonequivalent dependent variable [5] which would be affected by all infection control interventions except the MRSA universal screening. ITS analysis for both MRSA and MSSA series was carried out, Q3, 2007 as the intersection between segments. The null hypothesis was that changes in the positive blood cultures and positive systemic samples other than blood cultures for MRSA and the MSSA were similar.
We have included Durbin-Watson (DW) in the regressions statistics to correct for possible auto-correlation. DW value between 1.5 and 2.5 indicates that the values are independent. All analysis was done using SPSS 19.

Patients screened
The number of patients screened almost trebled from 1642 (3,432 specimens) in Q1, 2007 to 5087 (11,518 specimens) in Q2, 2007 then increased gradually over the next quarters to finally stabilize over 15000 (40,000 specimens) by Q1, 2009 (Figure 2). The ratio of patients to specimen numbers was about 1:2 in line with the swabbing of nose (N) and throat (T) only. However, this ratio has increased since Q2, 2008 to just under 1:3 when we introduced inpatient MRSA screening at every 10 days interval. The highest detection of on-admission prevalence of colonized patients was achieved (3.59%) in the Q3, 2007 then sharply reduced to 2%, Q2, 2008), 1%, Q3, 2010) and, finally, dipped down to 0.89% in Q4, 2010 (R 2 = 0.9), clearly reflecting the efficacy of decolonization and successful eradication (Figure 2).

Estimation of baseline prevalence of MRSA colonization
The highest rate of MRSA positive systemic samples (infections) other than blood cultures was 6.47% of all admissions in Q1, 2005 which implies a rate of colonization prevalence much higher than 6.47%. What was the on-admission colonization rate could not be extrapolated from inpatient infection burden but in Q1, 2007 the on-admission colonization detection rate was only 1% with risk based targeted screening. The highest on-admission detection rate of 3.59% was achieved in Q3, 2007 after universal screening was commenced (Figure 2). The impact of the universal screening and decolonisation on the MRSA burden was immediate: there was a sharp decline in on-admission colonisation prevalence from a peak at 3.59% in Q3, 2007 to less than 1% by Q4, 2010, while the number of patients screened remained stable at just over 15,000/quarter. These data reflect that detection by targeted screening before the universal screening was by far less efficient.

Impact on MRSA bacteremias
The target set required a reduction from 67 to 27 bacteremias by April, 2008 equivalent to numbers falling from 5.  (Table 2) revealed significant post-intervention drop of the baseline level of all MRSA bacteremias, representing an immediate effect, (β 2 -0.554, p 0.000) and the post-intervention declining trend (slope), representing a sustained effect (β 3 -0.393, p 0.048) where pre-intervention coefficient β 1 was close to 0 (−0.033,) signifying a flat trend. The immediate effect was equally prominent for hospital acquired MRSA bacteremias i.e., ≥48 h of hospitalization (β 2 -0.577, p 0.001) but post-intervention slope was not statistically significant (β 2 -0.216, p 0.298) as the numbers were already very low after the immediate sharp drop.
Impact estimation predicted 17 (all) and 13 (≥48 hours of hospitalization) MRSA bacteremias, respectively in the 19th quarter post-intervention if the intervention did not occur, and, predicted zero bacteremias with intervention which is borne out by the observation.
There was no statistically significant impact of universal screening on MSSA bacteremias. During the whole observation period of 34 consecutive quarters the number of MSSA bacteremias, both community onset (R 2 = 0.13) and ≥48 hours of hospitalization (R 2 = 0.09) have been declining slowly but not at a statistically significant level ( Figure 5, Table 2).

Impact on systemic Non-blood culture specimens
Similarly, there has been an equally significant impact on the number of MRSA isolates from non-blood culture systemic sources: sterile fluid, tissues, respiratory specimens and swabs, superficial and deep ( Figure 6). The number of positive specimens as percentage of admissions dropped from 3.32% in Q2, 2007 to 1.51% in Q3, 2007. The declining trend continued steadily over the following quarters to the lowest 0.25% in Q4, 2010 (R 2 = 8.9). The ITS analysis revealed a pre-intervention    The immediate decline and post-intervention declining trend were also significant for MRSA non-blood culture systemic specimens from general practitioners (GP): β 2 -0.402, p 0.000; β 3 -0.338, p 0.025 whereas pre-intervention declining trend was not significant (Table 2).
Of note, however, there was no concomitant reduction in MSSA isolates from non-blood culture systemic sources observed. Instead, there was a slow but statistically significant post-intervention increasing trend in the rate of hospital MSSA (R 2 = 7.3, Figure 7) confirmed by ITS analysis (β 3 1.477, p 0.000). The ITS analysis also revealed post-intervention increasing trend in the numbers of non-blood culture systemic MSSA from GP (β 3 1.20, p 0.034) although the predicted regression line was not significant (R 2 = 0.06).

Cost of screening
Each specimen was associated with a consumable cost of £0.62 for a negative screen and, additional, £2.16 for a positive specimen (identification and sensitivity). Estimated cost of screening was £24,800 per quarter plus £1080 based on the current quarterly positive number of about 500. The yearly cost of universal screening was just over £100,000 for laboratory consumables alone. In Q1, 2012 a combined NP Elution Swab (ESwab) in Liquid Amies medium was introduced reducing the number of specimens to half. The Eswab costs £0.45 more compared to the standard Amies swab, so, total consumable cost would be only marginally lower but specimen numbers halved.

Mupirocin resistance
The resistance to mupirocin (hospital isolates) marginally increased from 1.7% in 2004 to 2.3% in 2011.

Conclusion
In the era of multi-resistant pathogens, focusing hospital resources on a single antibiotic-resistant pathogen as a sole approach to infection control is criticized as inherently flawed [6]. Questions were also raised regarding the impact of universal screening on MRSA bacteremia rate given the sensitivity of the screening, lack of isolation rooms, the relatively low effectiveness of current decolonization techniques, staffing time and financial resources required vis-à-vis (cost-) effectiveness and the  real reduction in risk achieved. Caution is advised for more evidence before it becomes routine considering resource requirements, practical difficulties and consequences [7]. In hospitals with an MRSA on-admission prevalence of < 5% targeted rather than universal screening is advocated [8].
In our study, all interventions were targeted at both MRSA and MSSA except the screening and decolonization but decline in MSSA bacteremias over the 34 consecutive quarters of observation was not statistically significant, which served as a natural control (non-equivalent dependent variable) in the analysis. In contrast, a drastic drop and quarterly declining trend in MRSA bacteremias continued, reaching zero following the introduction of universal screening. In national context, during the first 2 years following setting of the national target in November 2004, there was minimal reduction; however, from September 2006 onwards, rates declined dramatically to reach a reported 57% reduction by April-June 2008 [9][10][11]. In our organization, dramatic decline in MRSA bacteremias and systemic infections was clearly linked to the introduction of universal screening and decolonization. This is consistent with reports that widespread uptake of decolonization has made the key additional contribution [11,12]. The effect of hand hygiene, identification of patients with MRSA infections or colonization and isolation interventions alone leads only to a gradual reduction in MRSA burden over many years [13]. The Pathfinder project [14], an implementation project to evaluate the impact of universal screening recommended by the NHS QIS HTA modeling [15], also found a temporal association between the initiation of universal screening and a significant reduction in colonization from 5.5% to 3.5% by month 12. However, no significant reduction in the rate of MRSA bacteremias and Surgical Site Infections (SSI) during the study was found. A re-worked model using observed parameters from the Pathfinder study projected that colonization prevalence could reduce over three to five years to low endemic levels (0.5-1.8%) with a chromogenic agar based universal screening strategy. Our study results support this projection.
In one neighbouring NHS Trust only seven additional day case patients in one month period would have been identified as MRSA carriers using the DoH universal screening compared with their targeted approach [16]. However, this Trust had already embarked on an expanded screening programme in three phases and needed to extend the screening only to day cases and elective admissions to comply with mandatory universal screening commenced in April, 2009. It is only to be expected that a successful screening and decolonization programme over a period would eventually lower the burden and a corresponding fall in MRSA detection rate.
In previous studies fluoroquinolone restriction was associated with decline in MRSA infections [17]. In our study MRSA rates dropped during the period when fluoroquinolone were used unrestricted albeit at a low level of average 12 DDD/100 bed-days per month. Our Fluroquinolone consumption more than halved to under 5 DDD/bed-days from June, 2008 which further dropped to under 2 from June, 2010. This low level of consumption may have positively impacted on our sustained low burden of MRSA infections.
Once low on-admission prevalence has been consolidated and sustained, a first line 'risk assessment' screening tool to reduce swabbing, laboratory costs and identify the proportion of high risk patients who could be preemptively isolated is a logical way forward. One study found that a simple three question Clinical Risk Assessment (CRA) could perform to a similar level to universal single site swabbing, but with considerably reduced resource implications [15]. The CRA has to be based on the local MRSA epidemiology, infection control practices and vulnerability of the patient population. The ITS methods used in this study have several advantages over other quasi-experimental studies, because they are less likely to be influenced by certain biases [18][19][20]. Cyclical effects and underlying increasing or decreasing secular trends may contribute to observed intervention effects. While looking at the effect of specific intervention, the analysis has to take into account the added effect of pre-existing or newly introduced measures. Finally, auto-correlated data means that adjacent data points can be more similar (positive auto-correlation) or dissimilar (negative autocorrelation), leading to under-or over-estimates of effect, respectively. Unlike the classical statistical methods that assume the observed data are independent random variables, time-series analysis takes into account the relationships existing between consecutive observations, a phenomenon known as auto-correlation. However, as with all non-experimental designs, causal inference from ITS designs is limited because it is impossible to rule out alternative explanations for observed changes in the time series.
Our experience with the precipitous decline in the burden of MRSA provides clear evidence of the efficacy of universal screening and decolonization. We achieved a very high level of compliance with screening uptake and decolonization. However, this success has to be considered in the context of the highly empowered weekly HCAI meeting in operation since May, 2007. This group have been ensuring continued compliance to MRSA screening, PVC care plan use and hand hygiene reviewing weekly audits and feeding back to the relevant staff with timely recommendation of appropriate measures backed by the support from the management.
It is noteworthy that our success was in the background of a declining trend in MRSA infections in England [11], the USA [21] and elsewhere during the period. If there is an epidemiological explanation, perhaps, an effect predicated on natural biological trends involving specific MRSA strains is a matter of great interest [11,22,23].
The study reveals one disturbing trend. The rate of non-blood culture systemic MSSA positive samples from the trust since Q2, 2007 is increasing. Also, the MSSA bacteremia trend appears to be decreasing only slowly and not at a significant level. This warrants specific actions to combat the MSSA burden.