Search

Despite perceptions that institutional review boards (IRBs) delay research, little is known about how long it takes to secure IRB approval. We retrospectively quantified IRB review times at 10 large Veterans Affairs (VA) IRBs.We collected IRB records pertaining to a stratified random sample of research protocols drawn from 10 of the 26 largest VA IRBs. Two independent analysts abstracted dates from the IRB records, from which we calculated overall and incremental review times. We used multivariable linear regression to assess variation in total and incremental review times by IRB and review level (i.e., exempt, expedited, or full board) and to identify potential targets for efforts to improve the efficiency and uniformity of the IRB review process.In a sample of 277 protocols, the mean review time was 112 d (95% confidence interval [CI]: 105-120). Compared with full-board reviews at IRB 1, average review times at IRBs 3, 8, 9, and 10 were 27 (95% CI: 6-48), 37 (95% CI: 11-63), 45 (95% CI: 20-69), and 24 (95% CI: 2-45) d shorter, and at IRB 6, times were 56 (95% CI: 28-84) d longer. Across all IRBs, expedited reviews were 44 (95% CI: 30-58) d shorter on average than were full-board reviews, with no significant difference between exempt and full-board reviews. However, after subtracting the time required for Research and Development Committee review, exempt reviews were 21 (95% CI: 1-41) d shorter on average than were full-board reviews.IRB review times differ significantly by IRB and review level. Few VA IRBs approach a consensus panel goal of 60 d for IRB review. The unexpectedly longer review times for exempt protocols in the VA can be attributed to time required for Research and Development Committee review. Prospective, routine collection of key time points in the IRB review process could inform IRB-specific initiatives for reducing VA IRB review times.

Keywords: Confidence level, PSD estimation, WELCH algorithm, Degree of freedom, Random error Abstract. Approach is proposed for determining the confidence level of the power spectral density (PSD) estimation with give Degree-of-freedom (D.O.F.) based on WELCH algorithm, a modified periodogram estimator for estimating PSD. A relationship between the random error, defined as a ratio of variance and mean value, and the D.O.F. is achieved firstly due to the analysis results of the statistic characteristic quantities. Then a quantitative relation between the confidence level, described as confidence probability and confidence interval, and the D.O.F. is established. Some examples are provided finally to demonstrate the proposed method. It is showed from the results that the PSD WELCH estimation with 128 D.O.F. have probability 68.4% only to ensure the true PSD within 0.88~1.2 times the estimation value.

Protocol performance issues are important in communication protocol design and network management, especially for those protocols which run in high-speed networking environments. An accurate performance modelling and evaluation approach is necessary to obtain reliable performance estimations and to improve system performance. Using queueing models (QMs) or finite state machines (FSMs) alone is difficult to achieve this goal because many aspects that affect performance are not taken into consideration by the model. This thesis proposes a new performance model called performance extended finite state machine (PEFSM). PEFSM makes use of the strengths of both QM and FSM. A PEFSM is based on an FSM which is extended to include time and probability. Furthermore, the transition time is refined and divided into two parts: the transition wait time and the transition service time. This allows the PEFSMs to integrate message arrival and queueing models which provide useful and essential information necessary for studying real world protocols. PEFSMs are classified into three categories based on the message arrival characteristics: synchronous PEFSMs (SyPEFSMs), asynchronous PEFSMs (AsPEF SMs) and hybrid PEFSMs (HyPEFSMs). While the hybrid model is the most useful and realistic model for communication protocols, the other two models are also presented for completeness, and as a way to explain the hybrid model. A method for computing performance metrics based on SyPEFSMs is given in the thesis. Two types of AsPEFSMs — AsPEFSM-α and AsPEFSM-β — and their performance evaluation methods are also presented. Then a class of HyPEFSM which is a hybrid model of SyPEFSM and AsPEFSM-α is introduced. The proposed performance evaluation method for this class of HyPEFSM is basically the combination of those for SyPEFSMs and AsPEFSM-αs. Our performance mod effing and evaluation approach has been applied to various examples, including the alternating bit protocol and multi-stage interconnection network (MIN). The performance evaluation method for PEFSMs makes use of stochastic pro cess and queueing theory. A new queueing property for an M/G/1. with multiple job classes and an analogous property for AsPEFSM-os have been discovered and proved. As a first step in improving system performance, the thesis defines software performance bottlenecks based on PEFSMs. Two bottleneck identification methods are proposed and tested. This thesis also proposes a testing method called t-test which in most cases is able to obtain the service times of invisible transitions when the protocol implementation under test is given as a black box. Transition service times are important parameters in FSM-based performance models. Studies in the past have usually assumed the transition times to be known a priori without discussing how they may be obtained. Simulations and measurement experiments have been conducted to validate the methodologies proposed in this thesis. The results are quite promising.