Limits...
Dynamic treatment effect (DTE) curves reveal the mode of action for standard and experimental cancer therapies.

Choudhury KR, Keir ST, Ashcraft KA, Boss MK, Dewhirst MW - Oncotarget (2015)

Bottom Line: The methodology doesn't presuppose any prior form for the treatment effect dynamics and is shown to give consistent estimates with missing data.Second, we demonstrate that a combination of temozolomide and an experimental therapy in a glioma PDX model yields an effect, similar to an additive version of the DTE curves for the mono-therapies, except that there is a 30 day delay in peak inhibition.We show that resulting DTE curves are flat.

View Article: PubMed Central - PubMed

Affiliation: Department of Biostatistics and Bioinformatics, Duke University Medical Center, NC, USA.

ABSTRACT
We present a method for estimating the empirical dynamic treatment effect (DTE) curves from tumor growth delay (TGD) studies. This improves on current common methods of TGD analysis, such as T/C ratio and doubling times, by providing a more detailed treatment effect and overcomes their lack of reproducibility. The methodology doesn't presuppose any prior form for the treatment effect dynamics and is shown to give consistent estimates with missing data. The method is illustrated by application to real data from TGD studies involving three types of therapy. Firstly, we demonstrate that radiotherapy induces a sharp peak in inhibition in a FaDu model. The height, duration and timing of the peak increase linearly with radiation dose. Second, we demonstrate that a combination of temozolomide and an experimental therapy in a glioma PDX model yields an effect, similar to an additive version of the DTE curves for the mono-therapies, except that there is a 30 day delay in peak inhibition. In the third study, we consider the DTE of anti-angiogenic therapy in glioma. We show that resulting DTE curves are flat. We discuss how features of the DTE curves should be interpreted and potentially used to improve therapy.

No MeSH data available.


Related in: MedlinePlus

a. Features of a treatment effect curve b. Example of the use of the EM algorithm for estimation with censored data, using a scaled Gaussian density as model for the treatment effect curve. The black line shows a ‘true’ growth curve, whereby the tumor shrinks considerably before growing again. Tumor volumes below 20 mm3 were not palpable and these observations are set to 20 mm3. The green line shows the estimated growth curve by fitting a smoothing spline using the recorded data. The EM algorithm (red line) imputes data values for observations below 20 mm3. Successive iterates (colored on a graduated scale from red-orange-yellow) from the EM algorithm improve upon the estimated growth curve to come quite close to the true value by the 1000th iteration.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4546495&req=5

Figure 3: a. Features of a treatment effect curve b. Example of the use of the EM algorithm for estimation with censored data, using a scaled Gaussian density as model for the treatment effect curve. The black line shows a ‘true’ growth curve, whereby the tumor shrinks considerably before growing again. Tumor volumes below 20 mm3 were not palpable and these observations are set to 20 mm3. The green line shows the estimated growth curve by fitting a smoothing spline using the recorded data. The EM algorithm (red line) imputes data values for observations below 20 mm3. Successive iterates (colored on a graduated scale from red-orange-yellow) from the EM algorithm improve upon the estimated growth curve to come quite close to the true value by the 1000th iteration.

Mentions: The rate of growth in the control group was 13%/day with an SD of 2%/day across animals. Animalwise dynamic treatment effect (DTE) curves were obtained using the non-parametric or model based methods, depending on the extent of missing data (Figure 5), and then averaged across animals. Average DTE curves for all treatments exhibit a single peak, with peak size appearing to increase with dose (Figure 6(b)). We also observe that although the treatment effect is substantially diminished by 30 days, growth is still inhibited beyond this point, i.e. the regrowth rate is slower than the rate of control growth, as previously demonstrated in [25]. Next we characterized the DTE curves in terms of their salient features, namely peak height, peak location, duration (Figure 3a) as well as mean AUC, which represents the area under the curve per day of observation. We found a significant increasing trend in the peak height with dose at 5%/day/Gy (p-value = 0.001), from 15%/day at 5 Gy to 40%/day at 10 Gy (Figure 6(c)). Similarly there was an increasing trend in peak location, with a delay of about 2days/Gy (p-value = 0.02) from 10 days at 5 Gy to 20 days at 10 Gy (Figure 6(d)). Duration ranged from 10 days at 5 Gy to 21 days at 10 Gy. The trend in duration was marginally significant at 2.2 days/Gy (p-value = 0.05), although the duration at 5 Gy appears to be lower (2.9 days) than the rest (Figure 6(e)). There was no significant trend in mean AUC (mean = 12.3%/day, SD = 3%/day, p-value for trend = 0.08), although the mean AUC at 5 and 6 Gy appears to be lower than the rest (Figure 6(f)).


Dynamic treatment effect (DTE) curves reveal the mode of action for standard and experimental cancer therapies.

Choudhury KR, Keir ST, Ashcraft KA, Boss MK, Dewhirst MW - Oncotarget (2015)

a. Features of a treatment effect curve b. Example of the use of the EM algorithm for estimation with censored data, using a scaled Gaussian density as model for the treatment effect curve. The black line shows a ‘true’ growth curve, whereby the tumor shrinks considerably before growing again. Tumor volumes below 20 mm3 were not palpable and these observations are set to 20 mm3. The green line shows the estimated growth curve by fitting a smoothing spline using the recorded data. The EM algorithm (red line) imputes data values for observations below 20 mm3. Successive iterates (colored on a graduated scale from red-orange-yellow) from the EM algorithm improve upon the estimated growth curve to come quite close to the true value by the 1000th iteration.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4546495&req=5

Figure 3: a. Features of a treatment effect curve b. Example of the use of the EM algorithm for estimation with censored data, using a scaled Gaussian density as model for the treatment effect curve. The black line shows a ‘true’ growth curve, whereby the tumor shrinks considerably before growing again. Tumor volumes below 20 mm3 were not palpable and these observations are set to 20 mm3. The green line shows the estimated growth curve by fitting a smoothing spline using the recorded data. The EM algorithm (red line) imputes data values for observations below 20 mm3. Successive iterates (colored on a graduated scale from red-orange-yellow) from the EM algorithm improve upon the estimated growth curve to come quite close to the true value by the 1000th iteration.
Mentions: The rate of growth in the control group was 13%/day with an SD of 2%/day across animals. Animalwise dynamic treatment effect (DTE) curves were obtained using the non-parametric or model based methods, depending on the extent of missing data (Figure 5), and then averaged across animals. Average DTE curves for all treatments exhibit a single peak, with peak size appearing to increase with dose (Figure 6(b)). We also observe that although the treatment effect is substantially diminished by 30 days, growth is still inhibited beyond this point, i.e. the regrowth rate is slower than the rate of control growth, as previously demonstrated in [25]. Next we characterized the DTE curves in terms of their salient features, namely peak height, peak location, duration (Figure 3a) as well as mean AUC, which represents the area under the curve per day of observation. We found a significant increasing trend in the peak height with dose at 5%/day/Gy (p-value = 0.001), from 15%/day at 5 Gy to 40%/day at 10 Gy (Figure 6(c)). Similarly there was an increasing trend in peak location, with a delay of about 2days/Gy (p-value = 0.02) from 10 days at 5 Gy to 20 days at 10 Gy (Figure 6(d)). Duration ranged from 10 days at 5 Gy to 21 days at 10 Gy. The trend in duration was marginally significant at 2.2 days/Gy (p-value = 0.05), although the duration at 5 Gy appears to be lower (2.9 days) than the rest (Figure 6(e)). There was no significant trend in mean AUC (mean = 12.3%/day, SD = 3%/day, p-value for trend = 0.08), although the mean AUC at 5 and 6 Gy appears to be lower than the rest (Figure 6(f)).

Bottom Line: The methodology doesn't presuppose any prior form for the treatment effect dynamics and is shown to give consistent estimates with missing data.Second, we demonstrate that a combination of temozolomide and an experimental therapy in a glioma PDX model yields an effect, similar to an additive version of the DTE curves for the mono-therapies, except that there is a 30 day delay in peak inhibition.We show that resulting DTE curves are flat.

View Article: PubMed Central - PubMed

Affiliation: Department of Biostatistics and Bioinformatics, Duke University Medical Center, NC, USA.

ABSTRACT
We present a method for estimating the empirical dynamic treatment effect (DTE) curves from tumor growth delay (TGD) studies. This improves on current common methods of TGD analysis, such as T/C ratio and doubling times, by providing a more detailed treatment effect and overcomes their lack of reproducibility. The methodology doesn't presuppose any prior form for the treatment effect dynamics and is shown to give consistent estimates with missing data. The method is illustrated by application to real data from TGD studies involving three types of therapy. Firstly, we demonstrate that radiotherapy induces a sharp peak in inhibition in a FaDu model. The height, duration and timing of the peak increase linearly with radiation dose. Second, we demonstrate that a combination of temozolomide and an experimental therapy in a glioma PDX model yields an effect, similar to an additive version of the DTE curves for the mono-therapies, except that there is a 30 day delay in peak inhibition. In the third study, we consider the DTE of anti-angiogenic therapy in glioma. We show that resulting DTE curves are flat. We discuss how features of the DTE curves should be interpreted and potentially used to improve therapy.

No MeSH data available.


Related in: MedlinePlus