The majority of the 16 studies identified through the search strategy were retrospective in design. Two reviews and two prospective studies were identified. Table summarises the characteristics and results from the 16 studies identified for this review.
Summary of included articles
The search strategy used to identify articles for this review did not identify the same articles in the latest review published by Ghantoji, Sail et al. [4
]. Two articles included in the review by Ghantoji, Sail et al. [4
] were not include in our review. Conversely, our study identified and included eight studies not used by Ghantoji, Sail et al. [4
]. The primary reason for both these discrepancies is that our review examined the prolongation of LOS, whereas the focus by Ghantoji, Sail et al. [4
] was economic cost . Similarly our review did not include two articles identified by the review conducted by Dubberke & Wertheimer (2009), but did identify a further 11 articles not used by Dubberke & Wertheimer (2009). The reasons for this are the same as those just previously described in addition to the inclusion of recent publications. Nine articles were common to both reviews. The review by Ghantoji, Sail et al. [4
] identified four articles not identified by Dubberke & Wertheimer (2009). Conversely, Dubberke & Wertheimer (2009) identified five articles not used by Ghantoji, Sail et al. [4
The manner in which participants were identified for the studies differed, with several studies using International Classification of Disease (ICD) codes to identify cases [16
]. The use of ICD codes to identify participants does have the potential to reduce sensitivity and specificity when identifying cases of CDI as coding data is likely to underestimate cases. In addition, coding practices can vary between hospitals, and therefore multi-centred studies have a greater potential for variation in sample selection. Furthermore, the timing of an episode of CDI cannot be determined by such an approach.
Excluding the reviews, only three of the remaining fourteen studies were undertaken in countries other than the United States. The systematic review examining the economic costs of CDI undertaken by Ghantoji, Sail et al. [4
] identified only four of thirteen articles from the United States. In the review undertaken by Dubberke and Wertheimer [22
], one Australian study undertaken was identified as having been published as a letter to the editor [23
The data collected in the various studies differed considerably. The majority of studies collected basic demographic data, such as age and gender. Some studies collected data about co morbidities and used a severity index such as the Charlson co morbidity index [18
]. Data collected about variables such as antibiotic exposure or other drug therapies were limited [25
Findings from all studies suggested that CDI contributes to a longer LOS in hospital. It was not possible to pool data because studies varied considerably in design, sampling and data analysis techniques. In studies that used a comparison between persons with CDI and those without, the difference in the LOS between the two groups ranged from 2.8days to 16.1days [24
]. These data suggest that CDI does play a role in increasing the LOS in hospital.
In a retrospective cohort of over 18 000 non-surgical patients hospitalised for more than 48 h, Dubberke et al. [24
] took a nested subset using a matched-pairs analysis and found that the increase in LOS that could be attributed to CDI was 2.8days. Controls were matched to cases by a propensity score developed for data analysis. Using logistic regression, patient-specific probabilities of developing CDI were developed. The median LOS was determined for cases and controls, with the various median pair-wise lengths of stay being compared by using the Wilcoxon signed-ranked test. Attributable LOS was determined by calculating the median pair-wise difference between CDI cases and the controls [24
]. As this study did not include surgical patients, it is possible that patients with severe CDI, those requiring colectomies, were excluded, leading to a potential bias. The use of a propensity score to match controls was used in an attempt to reduce any potential bias between controls and cases when determining CDI-attributable LOS.
A study undertaken by Lumpkins et al. [28
] suggested a considerably longer LOS then that reported by Dubberke et al. [24
]. In a prospective cohort study comprising of critically ill patients admitted to an intensive care unit, those with and without CDI were compared. A logistic regression model was used for data analysis. The mean hospital LOS was 15.9days greater in patients who developed CDI compared to those who did not (34.9days versus 19.0days, p=0.003). When cases were compared regarding antibiotic exposure, those with minimal exposure were found to have a shorter LOS in hospital, but data regarding all antibiotic exposure prior to admission, such as outpatients, were not obtained in this study [28
]. Such a finding would suggest that collecting data on antibiotic exposure is needed in future studies that employ a similar methodology.
The methods of data analysis varied, as shown in Table . In the majority of studies, a regression model was developed to determine the impact that CDI had on LOS [16
]. The studies did not collect data concerning the time of onset of CDI; therefore, it is not possible to exclude the possibility of reverse causality, in which longer lengths of hospitalisation may have increased the risk of CDI. The issues associated with controlling for a potential time-dependent bias caused by the LOS in hospital raises some significant concerns, which will now be discussed.