Oerter K E, Guardabasso V, Rodbard D
Comput Biomed Res. 1986 Apr;19(2):170-91. doi: 10.1016/0010-4809(86)90014-5.
We have developed a new computer program for detection of "peaks" in sequential hormone measurements in longitudinal studies of episodic hormone secretion. The program provides: (a) several statistically based approaches to the estimation of the random measurement error as a function of hormone level; (b) peak detection based on analysis of first derivatives with logic that has been optimized for asymmetrical peaks with exponential decays; (c) several approaches to the estimation of tolerances for the first and second derivatives; (d) a sensitive curve-fitting approach, to distinguish between upstrokes, exponential decays, and flat baselines; (e) ability to detect multiple overlapping peaks; (f) analysis of "robustness" by systematically varying the threshold around the most-likely value; (g) superimposition of detected peaks, to evaluate "average peak shape"; (h) analysis of the "decay rate," to obtain an estimate of the disappearance rate constant and half-life; (i) use of a "discrete deconvolution" approach, to solve for the apparent instantaneous rate of secretion, and provision of an error analysis to obtain estimates of the precision of these derived values; and (j) correlation with other relevant series as a means of cross validating. The program has been tested extensively on real and synthetic data, and appears to perform well. The frequency of "false positive" peaks can be held at any desired low level, and can be prevented from increasing as sampling frequency increases. The number of arbitrary assumptions, approximations, or thresholds is held to an absolute minimum. These methods are natural, logical, and follow from first principles of statistics.
我们开发了一种新的计算机程序，用于在间歇性激素分泌的纵向研究中检测连续激素测量中的“峰值”。该程序提供：(a) 几种基于统计学的方法，用于估计作为激素水平函数的随机测量误差；(b) 基于一阶导数分析的峰值检测，其逻辑已针对具有指数衰减的不对称峰值进行了优化；(c) 几种估计一阶和二阶导数容差的方法；(d) 一种灵敏的曲线拟合方法，用于区分上升段、指数衰减和平坦基线；(e) 检测多个重叠峰值的能力；(f) 通过系统地改变最可能值周围的阈值来分析“稳健性”；(g) 叠加检测到的峰值，以评估“平均峰值形状”；(h) 分析“衰减率”，以获得消失速率常数和半衰期的估计值；(i) 使用“离散反卷积”方法来求解表观瞬时分泌速率，并提供误差分析以获得这些派生值精度的估计；以及(j) 与其他相关系列进行相关性分析，作为交叉验证的一种手段。该程序已在真实数据和合成数据上进行了广泛测试，并且表现良好。“假阳性”峰值的频率可以保持在任何所需的低水平，并且可以防止其随着采样频率的增加而增加。任意假设、近似值或阈值的数量被保持在绝对最小值。这些方法是自然、合乎逻辑的，并且遵循统计学的第一原理。
