Search

The precision of a grading system based upon average score of 8 items for histological evaluation of malignancy of epidermoid carcinomas of the larynx was studied by letting 6 pathologists examine biopsies from 22 new cases. Exploratory statistical tools as well as a statistical framework (variance component models) for separating and evaluating the noise components are suggested. The order of magnitude of the resulting overall measurement uncertainty allowed discrimination between patients, though if a patient is graded by only one pathologist, only results as the extremes of the scale of malignancy should be trusted as possible therapeutical guidance. Based upon an analysis of the basic structure of the grading system the single noise components making up the measurement uncertainty were determined. Of greatest importance was noice created by disagreement between the pathologists in their ranking of the patients on the scale of malignancy. Furthermore the pathologist used the individual items differently, and certain items seemed to conform little to the scoring system. An attempt to minimize these two last components was performed; however neither calibration of the pathologists nor omission of the deviating items improved the precision significantly.

Birthweight-for-gestational age charts were based on the analysis of 3888 consecutive births in which the gestational age was estimated by measuring the fetal biparietal diameter before the 20th week of gestation. The data showed, in contrast to previous studies, a linear relation between gestational age and birthweight without inflection after term. The regression parameters showed a strongly significant difference between girls and boys. Thus, boys were 1.5% heavier than girls at 190 days gestation and 3.6% heavier at 300 days gestation. The linearity in the birthweight-for-gestational age charts could be due to the more reliable gestational age based on early ultrasound.

The value of estrogen and progesterone receptor (ER and PgR, respectively) determinations in predicting the recurrence-free survival (RFS) has been evaluated in a group of 807 node negative breast cancer patients. All of these patients are enrolled in the Danish Breast Cancer Cooperative Group (DBCG) 77-1a and 82-a protocols for low risk patients, and none of them have received systemic adjuvant therapy. At a median observation time of 50 months and in an evaluation of the total patient population as an entity, ER+ patients had only a marginally significant (P = 0.07) longer RFS than ER- patients while PgR+ patients experienced a significant advantage (P = 0.02). Among patients subgrouped according to menopausal status, both ER and PgR statuses were found to be significant prognostic factors for predicting RFS in the premenopausal women (less than 50 years) but not in peri- or postmenopausal women. Using Cox's multivariate analysis, nuclear pleomorphy was found to be the only significant prognostic variable, while the value of PgR status as a prognostic factor approached significance (P = 0.065). Although knowledge of ER status did not significantly improve distinction between patients with good and poor prognoses in the relatively small subgroup of premenopausal patients (n = 120) when PgR status was known, ER+PgR- patients have a lower risk of recurrence or death than ER-PgR- patients. Using a log-likelihood model, significant and distinct cut-off limits for the definition of receptor positivity were found for premenopausal patients: these were 5 fmol/mg cytosol protein for ER and 10 fmol/mg cytosol protein for PgR. These cut-off levels may reflect the ability of the ligand binding assay method used to discriminate between tissues with and without receptor proteins. Qualitative assessment of receptor status was as valuable as quantitative expression of receptor concentrations in predicting the RFS of the natural course of the disease among node negative premenopausal patients.

"The method of expected number of deaths is an integral part of standardization of vital rates, which is one of the oldest statistical techniques. The expected number of deaths was calculated in 18th century actuarial mathematics...but the method seems to have been forgotten, and was reinvented in connection with 19th century studies of geographical and occupational variations of mortality.... It is noted that standardization of rates is intimately connected to the study of relative mortality, and a short description of very recent developments in the methodology of that area is included." (SUMMARY IN FRE), (excerpt)

Meningeal involvement (MI) by non-Hodgkin's lymphoma (NHL) was seen in 38/602 patients (6.3%). In relation to histologic subtype the frequency of MI was: Follicular small cleaved and mixed cell 2/128 (1.6%), small lymphocytic and diffuse small cleaved cell 2/83 (2.4%), large cell and immunoblastic 13/295 (4.5%), small noncleaved cell 6/31 (19%), lymphoblastic 15/66 (23%). Risk factors that predict for MI were, besides histologic subtype, age under 40 yr, clinical stage IV, site of involvement (bone marrow, bone, skin gastrointestinal tract), and a poor response to initial therapy. In a Cox multivariate model encompassing the intermediate and high grade malignancy groups of the Working Formulation (WF), the 3 most important risk factors were histology, age, and stage. The estimated 1-yr probability of MI for combinations of the 3 risk factors was: 3 risk factors (61%), 2 risk factors (15-28%), 1 risk factor (4-8%), 0 risk factor (1.5%). At the diagnosis of MI, 84% of the patients had evidence of advanced systemic NHL, and the median survival after MI was 10 wk. CNS prophylaxis with whole-brain irradiation and intrathecal chemotherapy can only be recommended in patients with 2 or 3 risk factors. Improvement of the systemic chemotherapy might be the most important factor for prevention of MI in NHL.

It is noted that the bivariate exponential distribution introduced by Marshall and Olkin (1967, Journal of the American Statistical Association 62, 30-40) allows semiparametric generalizations along the lines of the Cox regression model for survival data. Partial likelihoods for the regression parameters may be derived (here illustrated by the use of the profile likelihood construction), and in most cases standard Cox regression model software may be applied for the analysis with minor modification of the input files. The study was initiated by data on occurrence of metastases from breast cancer. Metastases may occur at various sites, here grouped into ten categories, and simultaneous as well as consecutive occurrence at several sites in common. It is desired to identify and compare risk factors for development of metastases at each site, and we illustrate on some of these data that the above models may be useful for this purpose.

The widely used 'person-years method' of calculating expected mortality has been discussed recently by several authors. In studies where mortality is either lower or higher than the standard mortality of some reference population, the use of exposure to death as an estimator of the expected number of deaths will generally lead to bias, always exaggerating the difference between study and standard mortality. This bias is examined in a proportional hazards model. The recent suggestion by Hartz et al.1 of calculating the mortalities of individuals during their 'potential follow-up time' is claimed to be only rarely feasible in practice.

If the distribution of onset age of disease in a well-defined prevalent (cross-sectional) population of patients is known, disease incidence rates specific for age and calendar time period may be estimated, assuming known mortality rates and a closed population. This paper develops a method of estimation, illustrates this method on Danish diabetes data, and discusses its general applicability. The prevalent population of diabetic subjects in Fyn County on July 1, 1973 was ascertained from prescriptions, and information on disease onset was obtained from the patients' medical records. In this study only patients with onset of disease before or at age 30 years were studied. The mortality of diabetic subjects in Denmark was estimated from retrospective hospital data covering the period since 1933, and historical age-specific population sizes of Fyn County were obtained from census data. The incidence of diabetes increases with calendar time and with age for most cohorts. The variation with age for a fixed calendar year is more complicated, however, usually displaying a local maximum at about the age of puberty and a higher incidence at the upper end of the studied age range.

By means of a double-labelling experiment, circadian variations in the kinetic parameters of the S phase of the hamster cheek pouch epithelium were studied. The evaluation of the experiment included a recently developed correction for deviations from the strict pulse interpretation of the labelling technique. Pronounced circadian variations were found in S phase influx and efflux; the diurnal mean of both was estimated as 0.5%/hr, when based on measurements of all nucleated epithelial cells. Variations in S phase influx seem mainly responsible for the diurnal variation in cell proliferation, although diurnal variation in DNA synthesis rate, and thus in mean transit time, was also found. The increases in LI and influx were closely correlated and related to the beginning of the dark period. A circadian variation in cell number was also observed.

The recurrence-free survival rates (RFS) after one-stage mastectomy and partial axillary dissection in 1242 low risk breast cancer patients with invasive ductal carcinoma with or without residual cancer tissue (RCT) in the wall of the biopsy cavity were compared. RFS was significantly lower in patients with RCT (RCT-positive) whether premenopausal (n = 416) or postmenopausal (n = 826). By applying the Cox multivariate analysis on RCT and various known prognostic criteria, the incidence rates for RCT-positive patients relative to RCT-negative patients were estimated. The relative risk by RCT-positivity was in the order of 1.45, indicating that RCT is an independent risk factor contributing an increased risk of recurrence of about 45%.

Using the projecting method, measurements were made of the size of nuclei of liver cells in liver biopsies taken from ten patients with normal liver histology and normal liver function tests and from two patients with acromegaly. For the statistical analysis a parametric model of the distributions of the nuclear radii, the chi distributions, was used, making an estimation of the frequencies of di-, tetra- and octaploid nuclei possible. The distribution of the lengths of the nuclear radii was consistent with normal distributions corresponding to the different ploidy classes. Diploid nuclei constituted 84 to 99 per cent of the liver cell nuclei, and the frequency of polyploid nuclei increased with increasing age. Samples of nuclei within the same biopsy revealed significant differences in the mean radius of diploid nuclei and in the frequency of polyploid nuclei. This biological variation must be taken into account in the interpretation of karyometric data. One of the patients with acromegaly had a higher frequency of polyploid nuclei than seen in normal patients, and in both patients with acromegaly the size of diploid nuclei was large compared with normals. The frequency of binuclear nuclei was unrelated to age and frequency of polyploid nuclei, but males were found to have more binuclear liver cells than females.

In a prospective adjuvant trial including patients with primary operable breast cancer, invasive carcinomas of ductal type were subjected to histological malignancy grading. The parameters investigated were tubule formation, number of mitoses and cell pleomorphism. A Cox regression model for survival data was applied to evaluate the impact of the histological parameters on prognosis in 1809 patients with low-risk carcinomas. Cell pleomorphism proved superior to the other histologic tumor characteristics. It was found that low-risk invasive ductal carcinomas with severe cell pleomorphism had an excess recurrence intensity of 209% relative to carcinomas with no pleomorphism. It is therefore suggested that polymorphous invasive ductal breast carcinomas, other things being equal, should be regarded as high-risk tumors in future clinical trials. Finally it was found that the tripartite malignancy grading 1, 2 and 3 characterizing each of the histological parameters was not equidistant. Consequently, the traditional tripartite histologic scoring needs reconsideration.

Of 602 patients treated for non-Hodgkin's lymphomas, 9 developed overt acute nonlymphocytic leukemia or preleukemia with refractory cytopenia and cytogenetic abnormalities of the bone marrow. A Kaplan-Meier estimate of the cumulative probability of leukemic complications was 6.3 +/- 2.6% (mean +/- SE) 7 years after start of treatment. All 9 patients with leukemic complications belong to a major subgroup of 498 patients treated with alkylating agents, predominantly cyclophosphamide. The risk of leukemic complications in this subgroup was compared with the risk in 312 patients treated with other alkylating agents for Hodgkin's disease, and with the risk in 553 patients treated with dihydroxybusulfan for ovarian carcinoma. Cumulative 9-year risks were 8.0 +/- 3.3%, 12.8 +/- 3.5%, and 7.1 +/- 1.9%, respectively. The general risk of secondary leukemia after long-term treatment with alkylating agents ranges from 1% to 1.5% per year from 2 to at least 9 years after start of treatment.

Relative mortality in the period 1970-80 was studied among Danish men and women who were unemployed and employed on the day of the 1970 census. The study population consisted of the total labour force in the age range 20-64 on 9 November 1970--that is, about 2 million employed and 22,000 unemployed people. Relative mortality was analysed by a multiplicative hazard regression model (as a natural extension of the standardised mortality ratio) and a multiplicative regression model with extra-Poisson variation. A significantly increased death rate (40-50%) was found among the unemployed after adjusting for occupation, housing category, geographical region, and marital state. Analysis of five main causes of death showed increased mortality from all causes, but especially from suicide or accidents. In areas where the local unemployment rate was comparatively high the relative mortality among the unemployed was lower. The increased mortality among the unemployed was interpreted as a consequence of health related selection as well as increased susceptibility associated with the psychosocial stress of unemployment.

A method to estimate the intrauterine fetal weight by use of ultrasound measurements of the fetal biparietal diameter (BPD) and the abdominal diameter (AD) is presented. From a consecutive series of single pregnancies the 238 pregnancies (3% of the hospital population) with ultrasound measurements obtained within 4 days before delivery were used in the estimation of birth weight. In addition, the estimated formula was applied on a test material consisting of 100 similarly selected pregnancies from the same hospital. To evaluate the expected selection effects, the birth weight for gestational age in the study group was compared with birth weight for gestational age in the total population. The weight could be estimated as 0.0351 X AD1.65 X BPD0.69 X exp(0.00196 X gestational age). Thus, the actual birth weight was within 83-120% of the estimated weight (95% prediction limits), with a residual coefficient of variation of about 9%. The gestational age could be omitted without major influence on the weight prediction. When applying the formula on the test material, 70% of the actual weights deviated less than 10% from the estimated fetal weight, but a tendency towards a slight overestimation of the weight for light for gestational age infants was found.

In survival analysis, as in other branches of applied statistics, confirmatory analysis is often requested before an initial finding can be cleared of suspicion of being due just to mass significance. This paper points out that the special temporal structure of the classical clinical trial with staggered entry allows re-use of the initial survivors by left-truncating their remaining survival times. This procedure can often save a considerable amount of calendar time. The procedure is illustrated by data from the Danish Breast Cancer Cooperative Group.

In a 7-year longitudinal study of 40 normal boys, initially aged 8.6-11.7 years, 24-h urine samples were collected every 3 months and analysed for LH and testosterone as well as for the presence of spermatozoa (spermaturia). Spermarche (onset of the release of spermatozoa) was estimated on the basis of age at the first observed spermaturia. Physical examination, including measurements of height and sitting height and staging of pubic hair (Tanner stage) was performed every 6 months. Spermarche occurred early in puberty when the median pubic hair stage was 2.5. Urinary testosterone did not reach maximum levels until approximately 2 years after spermarche. Peak height velocity occurred when urinary testosterone levels were low, 1-2 years before adult levels of testosterone were attained. Our results support the proposition that low levels of testosterone have a predominantly growth-promoting effect, whilst higher concentrations have an inhibitory effect on height spurt.