Dose individualisation in oncology using chemotherapy-induced neutropenia: Example of docetaxel in non-small cell lung cancer patients.

AIMS: Chemotherapy-induced neutropenia has been associated with an increase in overall survival in non-small cell lung cancer patients. Therefore, neutrophil counts could be an interesting biomarker for drug efficacy as well as linked directly to toxicity. For drugs where neutropenia is dose limiting, neutrophil counts might be used for monitoring drug effect and for dosing optimisation. METHODS: The relationship between drug effect on the first cycle neutrophil counts and patient survival was explored in a Phase III clinical trial where patients with non-small cell lung cancer were treated with docetaxel. Once the association has been established, dosing optimisation was performed for patients with severe toxicities (neutropenia) without compromising drug efficacy (overall survival). RESULTS: After taking into account baseline prognostic factors, such as Eastern Cooperative Oncology Group performance status, smoking status, liver metastasis, tumour burden, neutrophil counts and albumin levels, a model-predicted drug effect on the first cycle neutrophil counts was strongly associated with patient survival (P = .005). Utilising this relationship in a dose optimisation algorithm, 194 out of 366 patients would have benefited from a dose reduction after the first cycle of docetaxel. The simulated 1-year survival probabilities associated with the original dose and the individualised dose were not different. CONCLUSION: The strong relationship between drug effect on the first cycle neutrophil counts and patient survival suggests that this variable could be used to individualise dosing, possibly without needing pharmacokinetic samples. The algorithm highlights that doses could be reduced in case of severe haematological toxicities without compromising drug efficacy.

Machine-Learning Assisted Screening of Correlated Covariates: Application to Clinical Data of Desipramine.

Stepwise covariate modeling (SCM) has a high computational burden and can select the wrong covariates. Machine learning (ML) has been proposed as a screening tool to improve the efficiency of covariate selection, but little is known about how to apply ML on actual clinical data. First, we simulated datasets based on clinical data to compare the performance of various ML and traditional pharmacometrics (PMX) techniques with and without accounting for highly-correlated covariates. This simulation step identified the ML algorithm and the number of top covariates to select when using the actual clinical data. A previously developed desipramine population-pharmacokinetic model was used to simulate virtual subjects. Fifteen covariates were considered with four having an effect included. Based on the F1 score (an accuracy measure), ridge regression was the most accurate ML technique on 200 simulated datasets (F1 score = 0.475 ± 0.231), a performance which almost doubled when highly-correlated covariates were accounted for (F1 score = 0.860 ± 0.158). These performances were better than forwards selection with SCM (F1 score = 0.251 ± 0.274 and 0.499 ± 0.381 without/with correlations respectively). In terms of computational cost, ridge regression (0.42 ± 0.07 seconds/simulated dataset, 1 thread) was ~20,000 times faster than SCM (2.30 ± 2.29 hours, 15 threads). On the clinical dataset, prescreening with the selected ML algorithm reduced SCM runtime by 42.86% (from 1.75 to 1.00 days) and produced the same final model as SCM only. In conclusion, we have demonstrated that accounting for highly-correlated covariates improves ML prescreening accuracy. The choice of ML method and the proportion of important covariates (unknown a priori) can be guided by simulations.

Investigating the impact of target lesion selection on drug effect evaluation and tumour growth rate determination using tumour growth inhibition models: Example of malignant pleural mesothelioma patients treated with cisplatin alone or in combination with pemetrexed.

In the last update of the RECIST criteria in 2009, it was proposed that the number of target lesions to be followed over time for response-to-treatment assessment be reduced from 10 to 5 lesions maximum, with up to 2 per organ. We explored the impact of reducing the number of target lesion on the assessment of drug effect in a randomised phase III clinical trial using a tumour growth inhibition (TGI) model. Tumour size measurements from 441 (out of 456) patients were used to build two datasets for which observations were the sum of longest diameters of all measurable lesions (ALL dataset) or following the RECIST 1.1 recommendations (R1.1 dataset). TGI models incorporating a categorical covariate for treatment group or a pharmacokinetic metric (i.e. dose; simulated area under the curve) were used to describe the longitudinal tumour size kinetics. Drug exposure was not superior to treatment group at describing drug effect. ALL and R1.1 individual estimates of drug effect appeared to be strongly correlated (r2=0.88). Including pharmacokinetic metrics in TGI models should be conducted carefully when no pharmacokinetic samples are available. Reducing the number of target lesion did not seem to compromise the determination of drug effect using TGI models.

Impact of tumour size measurement inter-operator variability on model-based drug effect evaluation.

PURPOSE: During oncology clinical trials, tumour size (TS) measurements are commonly used to monitor disease progression and to assess drug efficacy. We explored inter-operator variability within a subset of a phase III clinical trial conducted from August 1995 to February 1997 and its impact on drug effect evaluation using a tumour growth inhibition model. METHODS: One hundred twenty lesions were measured twice at each time point; once at the hospital and once at the centralised centre. A visual analysis was performed to identify trends within the profiles over time. Linear regression and relative error ratios were used to explore the inter-operator variability of raw TS measurements and model-based estimates. RESULTS: While correlation between patient-level estimates of drug effect was poor (r2 = 0.28), variability between the study-level estimates was much less affected (9%). CONCLUSIONS: The global evaluation of drug effect using modelling approaches might not be affected by inter-operator variability. However, the exploration of covariates for drug effect and the characterisation of an exposure-tumour shrinkage relationship seems limited by the high measurement variability that translates to a poor correlation of individual drug effect estimates. This might be addressed by the use of more precise computer-aided measurement methods.

Population Pharmacokinetic/ Pharmacodynamic Modelling of Auditory-Evoked Event-Related Potentials with Lorazepam.

Event-related potentials (ERPs) are commonly used in Neuroscience research, particularly the P3 waveform because it is associated with cognitive brain functions and is easily elicited by auditory or sensory inputs. ERPs are affected by drugs such as lorazepam, which increase the latency and decrease the amplitude of the P3 wave. In this study, auditory-evoked ERPs were generated in 13 older healthy volunteers using an oddball tone paradigm, after administration of single 0.5 and 2 mg doses of lorazepam. Population pharmacokinetics (PK)/pharmacodynamics (PD) models were developed using nonlinear mixed-effects methods in order to assess the effect of lorazepam on the latency and amplitude of the P3 waveforms. The PK/PD models showed that doses of 0.3 mg of lorazepam achieved approximately half of the maximum effect on the latency of the P3 waveform. For P3 amplitude, half the maximum effect was achieved with a dose of 1.2 mg of lorazepam. The PK/PD models also predicted an efficacious dose range of lorazepam, which was close to the recommended therapeutic range. The use of longitudinal P3 latency data allowed better predictions of the lorazepam efficacious dose range than P3 amplitude or aggregate exposure-response data, suggesting that latency could be a more sensitive parameter for drugs with similar mechanisms of action as lorazepam and that time course rather than single time-point ERP data should be collected. Overall, the results suggest that P3 ERP waveforms could be used as potential non-specific biomarkers for functional target engagement for drugs with brain activity, and PK/PD models can aid trial design and choice of doses for development of new drugs with ERP activity.

Pharmacokinetics and Pharmacodynamics-Based Mathematical Modeling Identifies an Optimal Protocol for Metronomic Chemotherapy.

Metronomic chemotherapy is usually associated with better tolerance than conventional chemotherapy, and encouraging response rates have been reported in various settings. However, clinical development of metronomic chemotherapy has been hampered by a number of limitations, including the vagueness of its definition and the resulting empiricism in protocol design. In this study, we developed a pharmacokinetic/pharmacodynamic mathematical model that identifies in silico the most effective administration schedule for gemcitabine monotherapy. This model is based upon four biological assumptions regarding the mechanisms of action of metronomic chemotherapy, resulting in a set of 6 minimally parameterized differential equations. Simulations identified daily 0.5-1 mg/kg gemcitabine as an optimal protocol to maximize antitumor efficacy. Both metronomic protocols (0.5 and 1 mg/kg/day for 28 days) were evaluated in chemoresistant neuroblastoma-bearing mice and compared with the standard MTD protocol (100 mg/kg once a week for 4 weeks). Systemic exposure to gemcitabine was 14 times lower in the metronomic groups compared with the standard group. Despite this, metronomic gemcitabine significantly inhibited tumor angiogenesis and reduced tumor perfusion and inflammation in vivo, while standard gemcitabine did not. Furthermore, metronomic gemcitabine yielded a 40%-50% decrease in tumor mass at the end of treatment as compared with control mice (P = 0.002; ANOVA on ranks with Dunn test), while standard gemcitabine failed to significantly reduce tumor growth. Stable disease was maintained in the metronomic groups for up to 2 months after treatment completion (67%-72% reduction in tumor growth at study conclusion, P < 0.001; ANOVA on ranks with Dunn test). Collectively, our results confirmed the superiority of metronomic protocols in chemoresistant tumors in vivoCancer Res; 77(17); 4723-33. ©2017 AACR.