Table 1
Mean variation due to contracting-out cleaning services vis-a-vis retaining them in house on MRSA incidence rate.

Incidence rate of MRSA infection
Bivariate association Adjusted models Propensity score matching Heckman selection model
Mean variation due to contracting-out cleaning services vis-a-vis retaining them in house 0.42* (0.09) 0.22
(0.09) 0.29***
(0.05) 0.26
(0.33)
p-value under the null hypothesis of no-selection bias – – – 0.71
Number of Trust-years 582 582 446 582
Notes: Source: Data from Hospital data from Patient Environment Action Teams (PEAT) dataset (from 2010 till 2012), Patient-Led Assessments of the Care Environment (PLACE) (2013–2015), ERIC (Estates Return Information Collection) (2010–2015), NHS Inpatient Survey (2010–2014), NHS Staff Survey (2010–2014), and Public Health for England (2010–2014). Robust SE clustered at Trust level for models 1 and 2 and bootstrapped SE-values in parentheses (250 replications), stratifying by type of cleaning service, for models 3, 4 and 5. Coefficients represent average variation in MRSA incidence rate between Trust which outsource their cleaning services and those which retain their cleaning services in house.The dependent variable represents the incidence of MRSA infection at Trust level. Trust are matched through Matching (model 3) and their distribution are aligned by region, number of beds, number of specialist sites, number of multi sites. After having aligned the distribution we regress, through a linear model, the dependent variable on the number of beds, average length of stay, regional and year dummies.

*p

Next, to adjust for differences due to potential observable confounding across hospitals, we estimated the association of outsourcing with MRSA, adjusting for hospital size, patient mix, and complexity. As shown Table 1, after correcting for these potentially confounding factors, we find that outsourcing is still associated with 0.22 more cases of MRSA bacteraemia per 100,000 bed-days (95% CI: 0.04 to 0.39, p-value = 0.01). Again, to translate our estimation into a measure that will be meaningful in the original framework, we estimate the level of MRSA infection in our two scenarios, setting all the other covariates at their median value. According to this model, while Trusts outsourcing cleaning will report a MRSA rate of 1.32 per 100,000 bed-days, their matched in house comparator will report an average rate of 1.10.

As an additional step, we matched hospitals within geographic regions of the UK and to the nearest-neighbour on size and complexity. It was not possible to match 34 of the 126 Trusts using this method (including 18 Trusts with in-house cleaning and 16 that outsourced it) because they were too different in size (in 18 cases) or complexity (in 12 cases) or in terms of propensity itself (based on the maximum permitted difference – i.e. the caliper – between observations) (4 cases), leaving a total of 92 matched Trusts (see web appendix Table 3, Table 3b for more details).

Table 3
Association of contracting out cleaning services on economic cost outcomes.

Cost per bed Staff per bed
Mean variation due to contracting-out cleaning services vis-a-vis retaining them in house -£236* (33.7) −0.01 p.*
(0.002)
Number of Trust-years 446 442
Notes: Source: Data from Hospital data from Patient Environment Action Teams (PEAT) dataset (from 2010 till 2012), Patient-Led Assessments of the Care Environment (PLACE) (2013–2015), ERIC (Estates Return Information Collection) (2010–2015), NHS Inpatient Survey (2010–2014), NHS Staff Survey (2010–2014), and Public Health for England (2010–2014). Bootstrapped SE-values in parentheses (250 replications), stratifying by type of cleaning service. Coefficients represent average variation in MRSA incidence rate between Trust which outsource their cleaning services and those which retain their cleaning services in house. The dependent variable represents: cost for cleaning (per-bed column 1, measured in £), staff employed for cleaning per-bed (column 2, measured in people per bed [p]).Trust are matched through Propensity Score Matching and their distribution are aligned by region, number of beds, number of specialist sites, number of multi sites. After having aligned the distribution we regress, through a linear model, the dependent variable on the number of beds, average length of stay, regional and year dummies.

*p

Table 1 further presents the results of the matched models. As anticipated, this yields a more precise estimate, with outsourcing now associated with 0.29 more cases of MRSA bacteraemia per 100,000 bed-days (95% CI: 0.17 to 0.37, p-value = 0.01).

Trusts outsourcing cleaning report an average rate of MRSA bacteraemia of 1.34 per 100,000 bed-days while their in-house counterparts report an average rate of 1.05 per 100,000 bed-days.

Finally, we implemented a Heckman selection model to assess the possibility of selection bias into outsourcing. We do not find clear evidence suggesting selection (IMR = 0.27, p = 0.38) (Table 1 column 4). The coefficient is not, however, statistically significant, mainly because standard errors tend to be large when the common support condition is not reached (Caliendo and Kopeinig, 2008).

Table 2-presents the estimation of the association between outsourcing of cleaning services on outcomes other than MRSA infection rates, adjusting the differences between in-house and outsourced cleaning procedure through propensity score matching, namely percentage of staff reporting ready access to hand-washing material (column 1), percentage of patients reporting excellent cleanliness for the bathroom they used (column 2). We present the results in terms of the average variation in MRSA incidence between Trusts which outsource their cleaning services and those which retain their cleaning services in house. The variation in percentage points is presented in web appendix table 5