Following database searching, 1236 studies were identified, of which thirty-three were duplicates. Of the remaining 1203 studies, 1003 studies were excluded following title and abstract screening. A further 176 studies were excluded after the full-text screening, leaving 24 studies in the systematic review (11-35). The full selection process is summarised in figure 1. Six of the included studies had data for two or more antimicrobials (12-17). Each of these data was treated as a separate study, resulting in 34 studies, as shown in table 1 below.
Characteristics of included studies
Out of the 24 studies included for qualitative analysis, twenty-three were cohort studies (11-32, 35), and one was classified as a randomised controlled trial because it used data from two randomised trials (18). Sixteen studies were set on single hospitals (13, 14, 16, 17, 21-24, 26-32, 35), four used primary care data (12, 13, 19, 20), two were population-based (15, 16), one multicentre hospital-based (14), and one multicentre randomised controlled study (18). Six of the studies were carried in the United Kingdom (12, 13, 19-22), four each for China (23-26) and the United States (27-30), two each for Canada (15, 16) and Japan (31, 32), and one each for Australia (33), France (14), Israel (34), Singapore (17), and South Korea (35). The randomised controlled trial (18) was multinational, including nine countries, namely Australia, Benin, Brazil, Canada, Ghana, Guinea, Indonesia, Saudi Arabia, and South Korea. Out of the thirty-four studies included for analysis, 31 reported acute kidney injury (AKI), three reported on drug-induced liver injury (DILI), and none reported on tissue injury. After considering the classes of antimicrobials used, nine, eight and four studies reported on AKI due to aminoglycosides, glycopeptides, and macrolides, respectively. Cephalosporins, penicillin, quinolone, and trimethoprim had a couple of studies, each reporting on AKI, while nitrofurantoin had a single study reporting on AKI. Three studies were on antituberculosis antimicrobials, and they reported on DILI. Table 1 below summarises the main characteristics of all the included studies.
Risk of bias
The Newcastle-Ottawa Scale (NOS) (5) with slight modifications was used to assess the risk of bias for each of the included observational studies. This tool is well validated and commonly used for assessing the risk of bias for observational studies included in a systematic review (36). In this systematic review, the modified tool is shown in appendix 1. Table 2 above summarises the risk of bias for each included observational study. All included cohort studies had a low risk of bias. The only randomised controlled trial included in this review was assessed for risk of bias using the Cochrane risk of bias version 2 (RoB2) tool (37), and it was low.
None of the included studies determined the relative risk of organ injury among older adults using prescribed antimicrobials as exposure of interest and no antimicrobial prescribed as the control. However, 27 studies expressed the absolute risk of organ injury among exposed older adults. Nine studies (14, 17, 21, 22, 27, 33, 34) reported absolute risk of AKI due to aminoglycoside exposure, eight studies (23-26, 28-31, 38) reported absolute risk of AKI due to glycopeptide exposure. Four studies reported the absolute risk of AKI due to macrolide exposure (15, 16). One study for penicillin (32) and trimethoprim (19) also reported an absolute risk of AKI due to exposure to the respective antimicrobial. Only three studies (18, 20, 35) reported an absolute risk of DILI due to antituberculosis antimicrobials. Meta-analyses were done on absolute risks for AKI among older adults who received aminoglycoside, glycopeptide, or macrolide antimicrobials, and the respective attributable risk percentages were determined.
A total of 1611 participants were exposed to aminoglycosides, and 257 of them were reported to have developed antimicrobial associated AKI across the nine studies. Figure 2 below, is the funnel plot of the included nine studies showing a high degree of homogeneity. The random-effects model was used to meta analyse the studies to establish the overall risk of AKI among older adults using aminoglycoside antimicrobials. Figure 3 below, is the forest plot of the meta-analysis. The overall absolute risk of AKI among older adults exposed to aminoglycosides was 15.1% (95% CI: 12.5% - 17.7%). This was significantly higher (p < 0.0001) than the average risk of AKI among adults of 18 years and above, following aminoglycoside antimicrobial exposure (10.5%; 95% CI: 10.1% – 10.8%) (39). Therefore, the attributable risk per cent of AKI among older adults exposed to aminoglycosides was 30.5% (95% CI: 6.6% - 54.4%).
A total of 23,431 participants were included in the eight studies that reported AKI due to glycopeptide exposure among older adults. Figure 4 (additional file 5, document 1) is a funnel plot showing the distribution of the studies when the standard error was plotted against the proportion of cases, and figure 5 (additional file 5, document 1) shows the forest plot before removing outliers, with an absolute risk of 21.4% (95% CI: 17.1% - 25.7%). After performing the leave-one-out meta-analysis, the overall absolute risk reduced significantly (p = 0.017) after excluding the study by Carreno et al. (28). As shown in figure 6 (additional file 5, document 1), excluding other studies did not significantly impact the overall absolute risk of AKI. Figure 7 (additional file 5, document 1) shows the funnel plot following the outlier's exclusion, and figure 8 shows the forest plot for a random-effects model for the studies after removing the outlier. In this systematic review, the overall absolute risk of AKI following the use of glycopeptides is, therefore, 19.1% (95% CI: 15.4% - 22.7%). There is no significant difference (p = 0.117) with the established risk of AKI among adults (18 years and above) on glycopeptide antimicrobial treatment (absolute risk = 18.7%; 95% CI: 15.6% - 21.7%)(40).
Only four studies reported on the risk of macrolide-associated AKI, but among them, a total of 294,350 participants were involved. The fixed-effects model was used to analyse these studies. Figure 9 is the forest plot summarising the combined outcome of these four studies. The overall risk of AKI among older adults exposed to macrolides was 0.3% (95% CI: 0.3%-0.3%).
Drug-induced liver injury (DILI)
Only three studies, with 404 participants, reported antimicrobial-associated DILI (18, 20, 35). In their study, Noh et al.(35) followed up 77 outpatients who were diagnosed and being treated for latent tuberculosis infection (LTBI), of which 14.3% (=11) had raised aminotransferases (the laboratory marker used to determine DILI) due to antituberculosis antimicrobials. All the participants were 65 years or above. Similarly, Bright-Thomas et al.(20) followed up 2070 adults (18 years and above) with active tuberculosis over thirty years (1981 to 2010). A subgroup of two hundred of these patients was 70 years old or more, hence included in this systematic review. Five per cent (n=10) of these patients had raised aminotransferases, an indication of antimicrobial associated DILI. On the other hand, Campbell et al. (18) performed a post hoc safety analysis using data from two parallel, open-label, randomised controlled trials in 17 healthcare facilities from countries across the globe. A subgroup (n=127) of 65 to 90-year-old adults were considered for this systematic review, of which 5.5% (n=7) had antimicrobial associated DILI. All three studies agreed that antimicrobial associated DILI increased with age.
No studies included in this systematic review investigated the relative risk or odds ratio of organ injury associated with broad-spectrum antimicrobials compared to narrow-spectrum antimicrobials prescribed to older adults. However, three studies in one research (12) and another separate study (19) compared the odds of AKI among older adults when empirically prescribed different antimicrobials for urinary tract infection (UTI) with empirical prescription of nitrofurantoin. In another four studies published in one research paper (13), the odds of AKI among older adults prescribed different antimicrobials following UTI were compared with the odds following amoxicillin prescription. Another five studies each had data on long term compared to short term treatment and an age group comparison of those between 65 and 79 years versus those 80 years and above.
1. Organ injury following empirical antimicrobial prescription for UTI.
Ahmed et al. (12) and Ahmed et al. (19) carried out several studies to determine the odds ratios of AKI when older adults have prescribed either cephalexin, co-amoxiclav, ciprofloxacin or trimethoprim when compared to nitrofurantoin for UTI. All exposure antimicrobials, except trimethoprim, had similar odds of causing UTI to nitrofurantoin among older adults. Trimethoprim was 1.9 times more likely to be associated with AKI when compared to nitrofurantoin (95% CI: 1.5 – 2.5). Crellin et al. (13) also reported the odds ratios of AKI among the older adults when cefalexin, ciprofloxacin, nitrofurantoin and trimethoprim were compared to amoxicillin. The odds of AKI associated with nitrofurantoin or cefalexin was not significantly different to that of amoxicillin among the older adults (OR = 0.89; 95% CI: 0.65 – 1.24, and OR = 1.01, 95% CI: 0.74 – 1.37, respectively). Ciprofloxacin was 1.48 (95% CI: 1.03 – 2.13) times, and trimethoprim was 1.72 (95% CI: 1.31 – 2.24) times more likely to be associated with AKI among older adults when compared to amoxicillin when prescribed for UTI.
2. Organ injury following treatment duration.
Among the five studies that included duration of treatment as part of their analysis, three of them (17, 33, 34) compared up to 7days of treatment with aminoglycosides versus more than 7 days, while the other two studies (29, 30) compared short and long term treatments with glycopeptides. In the study by Paterson et al. (33), there was a strong association of AKI among the patients who received an aminoglycoside for more than seven days, with a risk ratio of 7.5 (95% CI: 2.7-20.5, p = 0.0001), when compared to those treated for seven or fewer days. Contrary, in both of their studies, Ong et al.(17) and Raveh et al.(34) found no association of risk of AKI among those who received an aminoglycoside for more than 7 days, when compared to 7 or fewer days of treatment, with relative risks of 0.99 (95% CI: 0.58-1.68) and 1.35 (95% CI: 0.68-2.68) respectively. Similarly, Gyamlani et al.(29) and Hall et al. (30) did not find any association between the risk of AKI among those who received long term glycopeptide treatment compared to short term treatment.
3. Organ injury according to age group
Three of the five studies determined the association of AKI due to glycopeptide treatment among those at least 80 years when compared to those between 65 and 79 years (25, 28, 30). According to Pan et al.(25), there was a significant association of AKI among those at least 80 years old and on glycopeptide therapy when compared to those between 65 and 79 years of age, with a relative risk of 2.3 (95% CI: 1.5-3.8, p = 0.0004). However, the studies by Hall et al. (30) and Carreno et al.(28) did not show any significant association of older age with AKI due to glycopeptide therapy. Similarly, both Bright-Thomas et al.(20) and Noh et al.(35) did not find any significant association of older age with DILI due to antituberculosis antimicrobials, with respective relative risks of 1.52 (95% CI: 0.44-5.19) and 1.18 (95% CI: 0.18-7.75).
Heterogeneity and publication bias
The included aminoglycoside studies were tested for heterogeneity and the Q-test was 11.43 (p = 0.1787), and I2 was 29.98% (95% CI: 0.00% - 67.58%). Figure 2 (above) is the funnel plot of the included nine studies showing a central and uniform distribution of both and large around the mean. The Egger’s regression test intercept was -0.3626 (p = 0.7031).
The funnel plot for studies on glycopeptide exposure (figure 4, additional file 5, document 1) was skewed to the right, with one study possibly an outlier. Figure 5 (additional file 5, document 1) is the forest plot before sensitivity analysis, with an absolute risk of 21.4% (95% CI: 17.1% - 25.7%). After performing the leave-one-out meta-analysis, the overall absolute risk reduced significantly (p = 0.017) after excluding the study by Carreno et al. (16). As shown in figure 6 (additional file 5, document 1), excluding other studies did not significantly impact the overall absolute risk of AKI. The Q – test was 39.08 (p < 0.0001) and the I2 test was 84.65% (95% CI: 70.15% - 92.10%). The funnel plot (figure 7, additional file 5, document 1) was skewed, and Egger's regression test intercept was 2.2664 (p = 0.0340).