The effect of the intervention programs for HAI control in the Bayesian

generalized time series model

Based on the longitudinal follow-up for the HAI incident count, we estimated the effects of different intervention programs with Bayesian Poisson linear ARIMA model. Table 5.6.1 shows the estimated results with linear trend effect and

autoregressive order one. The regression coefficients for CDC/TJCHA and bundle

96

care were both statistically significantly less than zero, revealing a protective effect with relative risks (RRs) of 0.76 (95% Credible Interval (CI): 0.66-0.87)) and 0.62 (95% CI: 0.54-0.72) for CDC/TJCHA and bundle care, respectively. The linear trend effect was insignificant but the autoregressive order one was significant.

After adjusting for age and gender, it was still insignificant for linear trend effect and significant for the autoregressive order one. Patients aged less than 40 (RR=0.13, 95% CI: 0.12-0.14) and 40-59 years (RR=0.30, 95% CI: 0.29-0.32) had smaller risk for HAIs compared to those aged elder than 70 years. The age and gender-adjusted effects of intervention program were similar to those in the crude model, RR=0.76 (95% CI: 0.68-0.86) for CDC/TJCHA and 0.62 (95% CI: 0.55-0.70) for bundle care. The tracking plot for the MCMC models shows that a good

convergence for this model.

If we took a six-month period for the buffering time for the intervention taking effect, the effect of this lagged intervention effect is shown in Table 5.6.2. Although the regression coefficients of time trend and autoregressive order one move toward the null hypothesis, it was still no significant time trend effect but had significant autoregressive effect. For the lagged 6-month intervention effect, CDC/TJCHA (RR=0.67, 95% CI: 0.58-0.77) and bundle care (RR=0.65, 95% C: 0.57-0.74)

97

remained significant. Their counterparts in the age and gender-adjusted model were similar (RR=0.66 (95% CI: 0.58-0.76) for CDC/TJCHA and 0.65 (95% CI: 0.58-0.73) for bundle care).

When we considered both with and without time-lagged intervention in the same model, it was still insignificant for trend effect and significant for autoregressive order one regardless of whether age and gender were adjusted (Table 5.6.3). After adjusting for age and gender, the effects of both with and without lagged six-month CDC/TJCHA and bundle care were statistically effective. In this model, the RR of HAI for concurrent CDC/TJCHA and bundle care were 0.85 (95% CI: 0.73-0.98) and 0.60 (95% CI: 0.51-0.72), and the lagged six-month RR were 0.68 (95% CI: 0.58-0.80) and 0.63 (95% CI: 0.55-0.71). Note that from previous results, we found the results with and without adjustment for age and gender were similar, so that we presented only the results with age and gender adjustment thereafter.

When autoregressive order two was taken into account, both first- and second-order autoregressive pattern were statistically significant in the model for the effect of concurrent intervention programs for HAI. CDC/TJCHA and bundle care remained statistically significant (Table 5.6.4).

If there considered third-order autoregressive pattern, all three autoregressive

98

terms were statistically significant. Interestingly, linear trend effect became statistically significant. The concurrent intervention programs of CDC/TJCHA (RR=0.74, 95% CI: 0.63-0.87) and bundle care (RR=0.61, 95% CI: 0.52-0.71) remained statistically significant (Table 5.6.5). The tracking plots on the relevant parameters were shown in (Table 5.6.5a).

In a third-order autoregressive and first-order moving average model (Table 5.6.6), the moving-average term was not statistically significant. The results for intervention programs were similar to previous third-order autoregressive model.

When it was lagged six-month intervention programs were considered in the three-order autoregressive and linear trend effect time series model, the linear trend was not statistically significant. For the effect of intervention program, the lagged six-month intervention programs of CDC/TJCHA (RR=0.64, 95% CI: 0.54-0.75) and bundle care (RR=0.65, 95% CI: 0.55-0.76) were statistically significant (Table 5.6.7).

The results of the generalized time-series model considered both concurrent and lagged six-month intervention programs were shown in Table 5.6.8. In this model, linear trend effect was insignificant, but all three autoregressive order terms were statistically significant. As far as the intervention effects were concerned, the RR of HAI for concurrent CDC/TJCHA and bundle care were 0.83 (95% CI: 0.70-0.99) and

99

0.57 (95% CI: 0.46-0.70), and the lagged six-month RR were 0.65 (95% CI: 0.53-0.79) and 0.62 (95% CI: 0.53-0.73).

5.7 The effect of the intervention programs for HAI control in the Bayesian

generalized linear mixed ARIMA model

As far as the heterogeneity across infection site, department, and pathogens were concerned, a series of Bayesian GLIMMIX-AR(1) models were conducted. Table 5.7.1 shows the results of random effect on the infection site. It was shown that the

random effect term was highest for UTI, and followed by Bacteremia, pneumonia, and SSI. The standard deviation for the random effect term was 0.638 (0.245). The

results for age, gender, season, trend effect, and autoregressive order one were similar to their counterparts of the fixed effect model in previous sections. In this model, the RRs of HAI for CDC/TJCHA and bundle care were 0.75 (95% CI: 0.66-0.86) and 0.63 (95% CI: 0.56-0.71), respectively.

For the Bayesian GLIMMIX-AR(1) model with random effect on departments, the random effect term was highest for surgical department, followed by Departments

of nephrology, chest, and GI, and lowest in pediatric and ER. The standard deviation for the random effect term was 1.216 (0.266). In this model, the hygiene was

statistically significant (RR=0.90, 95% CI: 0.83-0.99). The RRs of HAI control for

100

CDC/TJCHA and bundle care were 0.74 (95% CI: 0.65-0.84) and 0.60 (95% CI: 0.54-0.69), respectively. (Table 5.7.2)

For the Bayesian GLIMMIX-AR(1) model with random effect on pathogens, the random effect term was highest for Gram-negative, followed by Gram-positive, fungi,

and lowest in anaerobic. The standard deviation for the random effect term was 1.449 (0.495). In this model, the hygiene was statistically significant (RR=0.91, 95% CI:

0.84-0.98). The RRs of HAI control for CDC/TJCHA and bundle care were 0.75 (95% CI: 0.67-0.84) and 0.62 (95% CI: 0.55-0.69), respectively. (Table 5.7.3) The interaction terms were further considered in the Bayesian GLIMMIX-AR model in order to exam the effect of intervention program for different infection sites, departments, and pathogens. For the random effect on infection site, we successfully obtained the posterior distribution of the effects of two intervention programs of interests, CDC/TJCHA and bundle care, for different infection sites (Table 5.7.4). The CDC/TJCHA and bundle care did not work on pneumonia. The most striking effect of CDC/TJCHA was for SSI (RR=0.63, 95% CI: 0.46-0.85) and bacteremia (RR=0.64, 95% CI: 0.52-0.80), and followed by UTI (RR=0.84, 95% CI: 0.70-1.00) (others was not in discussion). The bundle care had similar effect on SSI (RR=0.56, 95% CI: 0.44-0.72), UTI (RR=0.62, 95% CI: 0.44-0.72), and bacteremia (RR=0.63, 95% CI:

0.53-101 0.75).

Table 5.7.5 shows the results of interaction for intervention programs and

departments. The CDC/TJCHA and bundle care did not work in oncology department.

The most striking effect of CDC/TJCHA was for ER (RR=0.06, 95% CI: 0.00-0.56).

For GI, cardiovascular, chest, and neurology, the RRs of CDC/TJCHA on HAI control were around 0.6-0.7 (p<0.05). In surgical department, the CDC/TJCHA had

significant effect on HAI (RR=0.74, 95% CI: 0.62-0.90). The effects were not

statistically significant in the Department of Pediatric, Nephrology, and Infection. The RR was statistically larger than one in the Department of Oncology (RR=1.46, 95%:

1.02-2.07). The bundle care was most effective in the Department of Infection

(RR=0.24, 95% CI: 0.15-0.37), followed by Chest (RR=0.34, 95% CI: 0.26-0.46), ER (RR=0.43, 95% CI: 0.30-0.62), Cardiovascular (RR=0.49, 95% CI: 0.34-0.69),

Neurology (RR=0.56, 95% CI: 0.43-0.74), Nephrology (RR=0.53, 95% CI: 0.42-0.67), GI (RR=0.58, 95% CI: 0.45-0.73), and Surgical (RR=0.66, 95% CI: 0.56-0.77). Note that the RRs were statistically significantly larger than one in the Department of Pediatric (RR=0.45, 95% 0.04-0.85) and Oncology (RR=1.61, 95% CI: 1.24-2.09).

Table 5.7.6 shows the results of interaction for intervention programs and pathogens. The most striking effect of CDC/TJCHA was for anaerobic (RR=0.35,

102

95% CI: 0.13-0.78). It could reduce around 30% HAI for both Gram-positive and Gram-negative. The effect on fungi was small and not statistically significant

(RR=0.95, 95% CI: 0.76-1.17). For bundle care, the most striking effect was still for anaerobic (RR=0.18, 95% CI: 0.09-0.37). The second one was for fungi (RR=0.48, 95% CI: 0.39-0.58). For Gram-positive and negative, the RR of bundle care was 0.66 (95% CI: 0.56-0.78) and 0.69 (95% CI: 0.62-0.78), respectively.