In our patient population, the majority of patients receiving drug monitoring had hematologic malignancy and presumptive invasive aspergillosis, although fewer than half (n = 46; 43%) had proven or probable invasive fungal disease. Consistent with prior studies, there was high variability in initial drug levels, and those levels did not correlate with voriconazole dosage [8, 15, 24, 25]. The time between initiation of voriconazole and initial drug level was 11 days in our study. Despite the long duration between drug initiation and TDM, one-third of our patients had subtherapeutic initial drug levels. When we compared dosing regimens in patients with levels < 1.0 mg/L to those with levels ≥ 1.0 mg/L, we found no difference in the proportion who received intravenous formulations or weight-adjusted dosage, suggesting that weight and route of administration are not the primary determinants of drug levels [26, 27].
Furthermore, similar to prior studies, we found that the majority of patients who had subtherapeutic or supratherapeutic levels on initial testing had similar subsequent results. Because of the nature of our study, we do not have data regarding the presence of genetic polymorphisms that affect drug metabolism such as CYP2C19 in our population that could explain some of the variability in initial drug levels [15, 28, 29]. A further analysis of clinical and radiologic outcomes among the patients who were Caucasian, in whom the CYP2C19 polymorphism is more prevalent, did not reveal any difference in associations between drug level and outcome as compared to the total study population. It is also possible that for a subset of our patients, voriconazole drug levels are not reflective of true trough concentrations because of inappropriate timing of drug level monitoring relative to voriconazole dose administration. This may be particularly true among outpatients for whom we do not have information regarding dosage in relationship to drug level monitoring. In prior studies, there has been high variability in physician ordering practices of voriconazole levels, including random drug levels being checked rather than true trough levels . In our study, however, we found that the median time after dose administration was close to ideal at 11.3 hours. Nonadherence to therapy is also a factor that is associated with subtherapeutic drug levels, and has been examined in prior studies by measuring presence of voriconazole metabolites in patient blood samples . These factors are less likely to be the case in our study given clear documentation of the timing of dose administration and blood sample collection among our inpatients, though this data is often incomplete for our outpatient population.
No significant relationship between achievement of therapeutic drug levels and clinical response to therapy at 12 weeks or radiologic response was demonstrated among patients with proven or probable invasive fungal infection. Surprisingly, at 6 weeks, there was a significant association between subtherapeutic drug levels and clinical response. These results are in contrast to prior studies, which do show a relationship between therapeutic drug levels and clinical efficacy [7, 10, 11]. This discrepancy may be a result of limiting our analysis of clinical outcomes to patients with proven or probable IFIs. Prior studies have examined clinical response in all patients with voriconazole TDM, including those without proven or probable IFI . In addition, the timing of clinical outcome assessment may have been different from prior studies, as well as the sociodemographic profile of the patient population . We chose to perform our assessment at 6 and 12 weeks, rather than at 2 weeks, due to the prolonged period of immunosuppression and duration of antifungal therapy in many of our patients.
An important potential selection bias in our study was the selective inclusion of patients for whom voriconazole TDM was performed. There were many patients who received voriconazole who did not have their levels monitored, and we were unable to capture data on these patients. The decision to order voriconazole levels was at the discretion of the clinician. An algorithm was put into place for routine ordering of voriconazole levels, though this was not implemented until early 2009, and did not affect most of the patients whom we studied. We have, however, attempted to address the bias by performing the evaluation of clinical outcome only among patients with proven and probable invasive fungal infection, for whom initial voriconazole TDM was likely to be performed regardless of response to therapy. However, selection bias could potentially explain why we did not show a relationship between achievement of therapeutic drug levels and clinical response.
We were limited in our study by an inability to capture mortality data, given the retrospective nature of the chart review and the loss to follow-up among some inpatients who transitioned to outpatient care. However, for the majority of patients, we were able to obtain outcomes measured at 6 and 12 weeks after initiation of voriconazole therapy, and we have based our analysis on these endpoints. We also do not have information on the microbiologic isolates of the pathogens to establish the relationship between voriconazole serum concentration and pathogen MICs, as performed in prior studies because this was not routinely performed at our institution in 2007–2009 [32, 33]. We explored whether differences in confounders between patients with subtherapeutic and therapeutic levels could account for the difference in therapeutic outcome. We did not find differences between the two groups in weight, steroid use, incidence of graft-versus-host-disease, duration of therapy, or use of combination antifungal agents. We were unable to evaluate the effect of sepsis as a potential confounder because many of our patients had prolonged complicated hospitalizations related to their hematologic malignancies, and it was difficult to determine whether sepsis during their hospitalization had an effect on their voriconazole concentrations. We speculate that in our population with a high proportion of patients with hematologic malignancy and hematopoietic stem cell transplantation, host immune response is an important predictor of outcome.
There is currently no consensus on optimal timing for voriconazole drug monitoring. It is unclear whether routine voriconazole drug monitoring should be performed, or whether the clinical utility of TDM is limited to patients with adverse events. The study by Dolton et al. included all patients for whom trough levels were checked on day 2 of dosing in those with loading doses, and on day 7 or later in those without loading doses . All of our patients had drug monitoring performed at least 3 days after initiation of therapy, suggesting that they had reached a steady-state level by time of monitoring.
A high frequency of adverse drug events was seen in our patient population, with a higher incidence of acute renal failure in patients with supratherapeutic levels. Despite prior reports associating cyclodextrin in intravenous formulations with renal failure , we did not find a difference in prevalence of intravenous dose administration between patients who did and did not have renal failure. We also found no association between supratherapeutic levels and hepatotoxicity, in contrast to prior studies which have noted significantly higher levels of liver function test abnormalities with higher trough levels [11, 29]. Also, we did not find that the patients with hepatotoxicity received doses higher than the standard 4 mg/kg dosing, consistent with prior data showing that higher doses do not correspond to hepatoxicity . Similar to prior studies, a trend suggesting that supratherapeutic voriconazole levels were associated with encephalopathy was seen, though this was not a statistically significant association . It is possible that the nature of the chart review did not allow us to capture information regarding the presence of central nervous system effects because it relied on provider charting of symptoms. However, overall our population had rates of adverse events similar to prior studies suggesting that despite use of chart review for data extraction, we were able to capture a substantial proportion of the adverse events experienced by our patients .
Our results stand in contrast to a recent randomized controlled trial conducted at a single center in Korea showing improvement in clinical outcomes and decrease in discontinuation of therapy due to adverse events with therapeutic drug monitoring . We believe that this reflects both a difference in our patient population and the absence of a clinical trial design in our study. Participation in clinical trials may change the behavior of physicians and patients, and results are often difficult to replicate in clinical practice . Our study reflects the outcomes seen in practice at an academic center highly experienced with management of invasive fungal infections in immunocompromised hosts. We propose that further studies be performed to evaluate whether implementation of the clinical trial algorithm is efficacious in a clinical setting.