Just received this from another poster. Do not know if that person wants credit for finding it or not. Deals with your question. See the part about combining endpoints lowering the statistical power of each individual endpoint. I do not understand how or why but some other person may.
<<<<<<However, use of
combined morbidity and mortality observations as dual primary end points requires statistical
adjustments>>>>>>>>>
Monday, Aug 16 1999 11:40PM ET
OK. Point made: other companies reveal the data in summary form before filing. I'll stop with that for the
moment.
On to something new. Take a look at this site.
chestnet.org
It may be too late for us, or at least me if you've read the following before. It suggests another explanation for
Mr. Castello's reticence: maybe Xoma really hasn't finished figuring it all out yet.
<<Phase 3 Trials
In the case of severe sepsis in which 28-day mortality is typically 30 to 40%, mortality is a meaningful,
practical, and appropriate choice as a primary end point. The optimal time for measurement of mortality
reflects a balance in that early measurements may exclude sepsis- or
treatment-related deaths whereas late measurements may include nonsepsis-related deaths (those attributed to
preexisting conditions). Mortality at 28 days has been considered a reasonable choice
since mortality curves have an inflection point around 20 days and after 28 days, mortality rates are relatively
stable and largely reflect underlying disease.
In considering the use of survival curve analysis (ie, time to death) vs point (landmark) survival, a potential
confounding factor is that an intervention that decreases the risk of early (day 1 to 3) inflammatory deaths, but
increases later deaths, might result in a shift in survival curves with no beneficial effect on overall survival.
Changes in the proportion surviving at the end of the study are the most unambiguous measure of benefit.
Nonmortality measures (eg, organ function, time in ICU) are also clinically significant. However, use of
combined morbidity and mortality observations as dual primary end points requires statistical
adjustments that lower the power of the trial to detect benefits of either kind. Such outcomes should be
measured as secondary end points and may be of particular value in assessing overall drug utility
and in selecting candidate agents for combined therapy. For agents whose physiologic effects are well defined,
such as improvements of BP and oxygenation, measurements of these parameters alone might predict clinical
utility. However, most agents in development (eg, agents that interfere with the inflammatory response) have
the potential to cause harm via immunosuppression. Hence, laboratory indexes of the inflammatory state
(cytokine levels and other markers of inflammation)
should not be used as surrogate end points without extensive prior evaluation.
Mathematical and statistical modeling could be more widely applied to risk stratification at entry, defining more
homogeneous subgroups within sepsis populations, and measuring trends over time in the study. In
cooperative multisite studies, mathematical and statistical modeling is critical to assure consistency in
case-mix, overall observed-to-expected mortality rates of placebo and treatment
groups, and use of nonstandardized protocols regarding admission criteria to the ICU, antibiotic therapy, and
resuscitation.
Laboratory and clinical parameters to be included in risk prediction models should be ascertainable at the time
of entry so that randomization can be stratified. Techniques for model development
traditionally include multiple logistic regression and Bayesian logic. The specialized sepsis models described
above use logistic regression. For model performance, all use receiver operator characteristic curve for
discrimination and less standardized approaches to calibration. Because sepsis presents such complex clinical
scenarios, newer nonlinear approaches for dynamic modeling of physiologic processes, including neural
networks and chaos theory, are being pursued.
The general ICU severity models include the following: acute physiology and chronic health evaluation
(APACHE II/III), mortality probability model (MPM II), simplified acute physiology score model (SAPS II),
and organ system failure models. Organ system failure models are generally useful for detecting trends in
patients' progress, but have not been designed as probability models.
All of the general severity models were developed with a focus on the first 24 h after ICU admission. Since
patients may develop severe sepsis much later into the ICU course, or even after ICU discharge, these models
are often not applicable. The three general ICU models also do not show good calibration for the subgroup of
patients who have severe sepsis at, or shortly after, ICU admission.
Models better suited for severe sepsis are needed since the general ICU models are not well adapted for severe
sepsis trials owing to differences in (1) study duration, (2) lead time bias, and (3)
necessity of approximations. In response to this need, specialized models for severe sepsis were modified
from existing databases and include the following: APACHE III sepsis model, MPM II customized for sepsis,
SAPS II customized for sepsis, and Study to Understand Prognoses and Preferences for Outcomes and Risks
of Treatment (SUPPORT) model for patients with multiorgan failure and sepsis. Most of these models are not
externally validated (the APACHE III sepsis model is). They do provide a 28-day mortality estimate. For
patients admitted to ICU with severe sepsis, MPM II sepsis, SAPS II sepsis, APACHE III sepsis may be
useful predictors of outcome.
Mediator measurements have intrinsic appeal because they have the potential to reflect underlying events in a
biologically and immunologically meaningful fashion. Proposed roles for mediator
measurements in severity models include risk stratification at entry, defining more homogeneous groups
within sepsis categories, improving prediction of mortality in severity models, and targeting therapies to
specific patient populations. However, the validity of mediator measurements in peripheral blood has not been
established and reliance on blood determinations does not take into account the potentially greater importance
of tissue and organ levels. For use in prospective risk stratification, for targeted therapy, or as an entry
criterion, such a test would have to be rapid and highly reproducible. Furthermore, if mediator levels are to be
used as a guide to therapy, we need more complete information on temporal changes in these levels related to
the natural history and pathogenesis of sepsis, resolution of inflammation and injury, and the impact of
underlying disorders
on baseline levels. At present, information is lacking to ascribe qualifiers of "benefit" or "harm" to a particular
cytokine profile.
While current therapies for severe sepsis are focused on selective downregulation of immune responses,
emerging data suggest that many patients are already immunosuppressed or may become functionally
immunosuppressed due to compensatory downregulation of the inflammatory response. Thus, an alternative
strategy is to use agents that enhance or restore immune responses of severely infected patients. Several
approaches to enhance host inflammatory responses have been evaluated, including interferon-gamma,
PGG-glucan, and granulocyte colony-stimulating factor, and more work is needed in this area.
A shortcoming of existing models is that they focus on the actions of a single mediator and fail to account for
the redundancy and interdependence of inflammatory responses. Clinical experience and
the complexity of host defenses suggest that no single agent will result in more than a 15 to 20% improvement
in mortality in severe sepsis. Hence, the development of a multimodal approach directed at various steps has
theoretical appeal to correct abnormalities or imbalances in the host inflammatory response. However, there is
scant information at present to direct the timing, dose, duration, or the specific components of combination
therapies.>> |
