When are summary ROC curves appropriate for diagnostic meta-analyses?

Diagnostic tests are increasingly evaluated with systematic reviews and this has lead to the recent developments of statistical methods to analyse such data. The most commonly used method is the summary receiver operating characteristic (SROC) curve, which can be fitted with a non-linear bivariate random-effects model. This paper focuses on the practical problems of interpreting and presenting data from such analyses. First, many meta-analyses may be underpowered to obtain reliable estimates of the SROC parameters. Second, the SROC model may be inappropriate. In these situations, a summary with two univariate meta-analyses of the true and false positive rates (TPRs and FPRs) may be more appropriate. We characterize the type of problems that can occur in fitting these models and present an algorithm to guide the analyst of such studies, with illustrations from analyses of published data. A set of R functions, freely available to perform these analyses, can be downloaded from (www.diagmeta.info).

Impact of a theoretically based sex education programme (SHARE) delivered by teachers on NHS registered conceptions and terminations: final results of cluster randomised trial.

OBJECTIVE: To assess the impact of a theoretically based sex education programme (SHARE) delivered by teachers compared with conventional education in terms of conceptions and terminations registered by the NHS. DESIGN: Follow-up of cluster randomised trial 4.5 years after intervention. SETTING: NHS records of women who had attended 25 secondary schools in east Scotland. PARTICIPANTS: 4196 women (99.5% of those eligible). INTERVENTION: SHARE programme (intervention group) v existing sex education (control group). MAIN OUTCOME MEASURE: NHS recorded conceptions and terminations for the achieved sample linked at age 20. RESULTS: In an "intention to treat" analysis there were no significant differences between the groups in registered conceptions per 1000 pupils (300 SHARE v 274 control; difference 26, 95% confidence interval -33 to 86) and terminations per 1000 pupils (127 v 112; difference 15, -13 to 42) between ages 16 and 20. CONCLUSIONS: This specially designed sex education programme did not reduce conceptions or terminations by age 20 compared with conventional provision. The lack of effect was not due to quality of delivery. Enhancing teacher led school sex education beyond conventional provision in eastern Scotland is unlikely to reduce terminations in teenagers. TRIAL REGISTRATION: ISRCTN48719575 [controlled-trials.com].

Designing follow-up intervals.

When an examination is needed to determine if an event has occurred there will be a loss of efficiency in using the resulting interval-censored data instead of the exact event times. In designing follow-up intervals this loss for longer intervals needs to be weighed against extra visits required by shorter intervals. We obtain results to quantify this for the estimation of the median and mean survival and for covariates in parametric regression models with equally spaced examination times. Asymptotic information loss for the log-normal and Weibull distributions are similar when comparisons are made between corresponding members of the two families. For distributions with coefficients of variation of 50 per cent or more, a choice of interval from 0.25 to 0.7 times the median is recommended.

Random effects models for HIV marker data: practical approaches with currently available software.

The analysis of marker data from HIV positive patients has been the motivation for many new developments in applied statistics. As well as reviewing these methods, this paper considers the extent to which programs to implement them are available in current software. Particular areas of development have been the joint modelling of markers and survival outcomes, non-linear random effects models that are of particular relevance for studying the efficacy of treatments and the use of Bayesian computational methods for inference from marker data. The package WinBUGS is recommended as being particularly well suited to the analysis of marker data.

Effect of combination therapy on immunologic progression of human immunodeficiency virus at a population level.

There is accumulating evidence from clinical trials and cohort studies that highly active antiretroviral combination therapy is effective at halting immunologic and clinical progression of human immunodeficiency virus (HIV). Its impact at a population level is less well known because the regimes may be difficult to tolerate and compliance poorer. The authors make use of population data for almost all of the HIV-infected people in Scotland in 1997 who were under clinical care and monitor their response to therapy during the first year when these effective treatments became widely available. More than two thirds of the HIV-positive patients were on some form of antiretroviral therapy during the year. The authors show that all treated groups, even those who were on changing regimes, showed net improvement in immunologic status during the year. For the group of patients on triple or quadruple therapy, there was an average increase of more than 100 CD4 cells/mm(3) over the year, with other treatment groups showing more modest, but significant, increases.

Balance in cluster randomized trials.

This paper explores the role of balancing covariates between treatment groups in the design of cluster randomized trials. General expressions are obtained for two criteria to evaluate designs for parallel group studies with two treatments. The first is the variance of the estimated treatment effect and the second is the extent to which the estimated treatment effect is changed by adjusting for covariates. It is argued that the second of these is more important for cluster randomized trials. Methods of obtaining balanced designs from covariates which are available at the start of a study are proposed. An imbalance measure is used to compare the extent to which designs balance important covariates between the arms of a trial. Several approaches to selecting a well balanced design are possible. A method that randomly selects one member from the class of designs with acceptable bias will allow randomization inference as well as model-based inference. The methods are illustrated with data from a trial of school-based sex education.

A single measurement of CD38CD8 cells in HIV+, long-term surviving injecting drug users distinguishes those who will progress to AIDS from those who will remain stable.

This study compares the predictive power of a single measurement of CD8+CD38+, CD8+CD45RO+ or CD8+CD38+CD45RO+ subpopulations in predicting progression to AIDS in a cohort of HIV+ long-term surviving injecting drug users. The results showed that both the total CD8+ percentage, and the CD8+CD38+ and CD8+CD38+CD45RO+ subpopulations of cells all individually predicted progression to AIDS. In combination with CD4, only the CD8+CD38+ subpopulation enhanced the predictive power of the CD4 percentage alone. The CD8+ percentage correlated negatively with the CD4 percentage and the CD8+CD45RO+ subpopulation did not predict disease progression. The proportion of CD8+CD38+ cells identified which patients with a moderate CD4 level were more likely to progress to AIDS, and conversely, which patients with a low CD4 count were likely to remain clinically stable. The results were consistent irrespective of whether time was measured from the date of seroconversion, or from the date of the test. This study is the first to measure these markers in HIV-infected injecting drug users, and in long-term survivors. The results demonstrate the considerable added value of the CD8+CD38+ cell percentage over the CD4 count alone, in predicting HIV clinical progression.

How to select covariates to include in the analysis of a clinical trial.

The comparisons of treatments in randomized clinical trials may use the analysis of covariance to adjust for patient characteristics. We present theoretical results that describe when such an adjustment would be expected to be beneficial. A distinction is made between covariates that are balanced in the design and those that are assigned by the randomization process. The results support the commonly held view that features balanced in the design of the trial (e.g., by stratification) and those that are strongly predictive of the outcome, and thus considered clinically prognostic, should normally be included in the analysis. For other covariates that are not balanced in the design, the potential benefits of including them in the analysis will depend on the number of patients in the trial. However, there is frequently a set of variables whose relevance is unknown and for which data-dependent methods of selection, based on the data for the current trial, have been proposed. A review of the literature has shown that these methods can produce misleading inferences. The decision as to which covariates to include in the analysis should be specified in the protocol on the basis of data from previous trials on similar patient populations. The methods are illustrated with data from a trial comparing two therapies for treating scalp psoriasis where the clinical importance of patients' age and sex as prognostic factors for efficacy is unknown. We show for what size of future trials it would be beneficial to adjust for these covariates and for what size trials it would not. In all cases, prespecification of variables to be included in the analysis is essential in order to avoid bias.

Standardisation and quality assurance of lymphocyte proliferation assays for use in the assessment of immune function. European Concerted Action on Immunological and Virological Markers of HIV Disease Progression.

Lymphocyte proliferation is a widely used technique to assess immune competence. However, the technique is subject to a large degree of variation, some biological and some technical. In this study, the components of variation in whole blood proliferation assays were analysed over time, using both antibody and mitogenic stimulants. The levels of variation within individual samples, between individuals and between groups of individuals over time were examined. A method of transforming the data is proposed which reduces the coefficients of variation to an acceptable level, and which expresses individual results as a standardised count. This method overcomes the problem of different levels of absolute counts, it corrects for time sensitive errors and allows data from multiple laboratories to be pooled.

The impact of British and US guidelines for initiating combination antiretroviral therapy on estimating national treatment requirements for HIV-positive patients. Collaborating Immunologists and the Scottish HIV and AIDS Group.

This paper uses comprehensive national data on HIV positive patients in Scotland to carry out a needs assessment exercise for combination antiretroviral therapy. The objective of this study was to estimate the numbers of HIV positive patients in Scotland who would be eligible for combination antiretroviral therapy under current British and US guidelines and to demonstrate the impact of these different guidelines on the resources required. The proportion of the Scottish population that would be eligible for combination therapy ranged from 76% to 91%, under different guidelines for initiating therapy. We thus estimate that for countries such as Scotland, including western Europe and the United States, where a large proportion of the HIV population became infected in the early to mid-1980s, the majority of patients will be eligible for combination therapy, regardless of the guideline.

Trends in incidence of and mortality from invasive cancer of the uterine cervix in Scotland (1975-1994).

OBJECTIVE: I. To identify major trends in the incidence of and mortality from invasive cancer of the cervix uteri in Scotland during the twenty year period 1975-1994; II. to consider the extent to which these trends may have been shaped by the introduction of systematic cervical screening. DESIGN: Analysis of annual age standardised and age specific rates for incidence and mortality, based on data collected by the Scottish Cancer Registry and the General Register Office for Scotland. SETTING: Scotland. SUBJECTS: Women registered with the Scottish Cancer Registry as having developed invasive cancer of the cervix during the period of interest. RESULTS: Annual all ages incidence rates of invasive cervical cancer show little overall change over the period 1975-1989, but exhibit a pronounced decline from 1990 onwards. All-ages mortality rates show clear evidence of decline during the period 1975-1994, the rate for 1994 being some 30% lower than that for 1975. Annual age-specific incidence rates show different patterns by age group, with clear evidence of decreasing trends in the age range 50-64 years but different patterns in younger and older age groups. Most age groups show steep declines in incidence from 1990 onwards. Age specific mortality rates for 1975-1994 exhibit the most pronounced decreasing trends in the age range 50-64 years. The trends identified are broadly similar to those experienced in England and Wales over an approximately comparable period. CONCLUSIONS: The overall (all ages) incidence of invasive cervical cancer in Scotland changed little during the period 1975-1989, but declined sharply from 1990 onwards. The most pronounced decline in incidence across the period 1975-1994 appears to have taken place in the age range 50-64 years. This decline has been accompanied by a commensurate fall in mortality in the same age range. These reductions in incidence and mortality may be attributable in part to increased coverage of cervical screening programmes during the period of interest. Evidence from other studies suggest that, without the increased coverage of cervical screening achieved during this period incidence rates in Scotland might have been seen to increase.

Modelling human immunodeficiency virus infection and acquired immune deficiency syndrome cases in Scotland: data sources, prior information and Bayesian estimation.

"The paper describes the methodology developed to carry out predictions of the acquired immune deficiency syndrome (AIDS) epidemic in Scotland. Information on the human immunodeficiency virus (HIV) epidemic comes from formal case reports of AIDS cases and HIV positive tests, reports from surveillance schemes and from special studies. These sources of information, up to the end of 1994, are reviewed.... A simple model of the HIV epidemic in Scotland is proposed and the information is summarized in terms of this model. Bayesian methodology, using Markov chain Monte Carlo methods, is described and used to predict future cases of AIDS in Scotland and people who will be living with AIDS in the years 1995-1999."

A comparison of different approaches for fitting centile curves to birthweight data.

Centile charts, plotted by week of gestational age, are widely used clinically to decide whether a child is 'small-for-dates', indicating that there may have been some cause of intra-uterine growth retardation. A variety of different statistical techniques have been proposed for constructing centile charts. Some methods are distribution free and work on the basis of obtaining and then smoothing the empirical centiles. In others the centiles are calculated assuming a Normal distribution either of the raw data or of transformed data. In Scotland the details of each maternity admission are entered onto an SMR2 form and these records are stored centrally by the Scottish Health Service. A data set consisting of all live births occurring in Scottish hospitals between 1980 and 1992 is used to compare different approaches to constructing centile charts.

The effect of human immunodeficiency virus infection and drug use on birth characteristics.

OBJECTIVE: To explore the effect of human immunodeficiency virus (HIV) infection and drug use on birth weight, length, and gestational duration at delivery. METHODS: Subjects had a history of injection drug use or a sexual partner who was an injection drug user, were Scottish, and their HIV serostatus during pregnancy was known. Control pregnancies were matched for age, parity, ethnic group, year of delivery, and postal code sector of home address. In addition, some were matched for smoking and housing deprivation score. Birth weights were standardized for gestational age by expressing them as z scores with a mean of zero and a standard deviation of unity. Statistical analysis was by univariate and multiple regression with multilevel modeling. RESULTS: Regression analysis for birth weight, gestational age, and gestation-adjusted birth weights (z score) included 789 pregnancies in 693 women. Human immunodeficiency virus seropositivity was associated with a z score that was 0.27 lower (P = .03), but there was no significant difference in gestational duration at delivery. Current oral or injection drug use were associated with a reduction in standardized birth weight (z score -0.27, P = .06, and z score -0.28, P = .04, respectively), and injection drug use with a reduction in occipitofrontal circumference only (1.8 cm reduction, P = .05). Injection drug use, but not the other factors, had an effect on gestational age at delivery (1.54 weeks earlier, P < .001). CONCLUSION: Although HIV seropositivity is associated with a small reduction in standardized birth weight, this effect is less than that attributable to smoking and may not be of clinical significance. The effect seems to be associated with placental size. Opiate use, regardless of route, had a small association with reduced birth weight, suggesting a specific drug effect. However, only injection drug use had a strong association with early delivery, and this effect was likely to be clinically significant at the population level.

CD4 and CD8 lymphocytes in diagnosis and disease progression of pediatric HIV infection.

Vertical infection with human immunodeficiency virus-1 (HIV-1) causes profound changes in the proportions of subpopulations of lymphocytes in the peripheral circulation. In this study the percentages in whole blood of CD4 and CD8 cells, and of immunologically important subpopulations, were measured in 19 HIV-infected children over periods of up to 4 years and compared to our recently published ranges for normal children of various ages. The rate of CD4 decline and of CD8 increase differed between clinically fast and slow progressors. On CD8 cells, cytotoxic, memory (CD11abright and CD45R0), and activation (HLA-DR) markers were raised soon after birth to levels outside the normal range, and compared favorably with HIV culture as a method for early diagnosis of HIV infection. Mean levels of naive (CD45RA) and memory (CD45R0, CD29) markers on CD4 cells became significantly altered after 48 months of age, suggesting that these are markers of more advanced disease. Despite different ages of enrollment into the study, in the cohort as a whole, the levels of the lymphocyte subpopulations studied changed consistently. Thus, their measurement could be useful both in the diagnosis and prognosis of HIV infection in individual children. This is the first report showing that lymphocyte subpopulation analysis can play a major role in the diagnosis of pediatric HIV infection.

Maternal HIV infection, drug use, and growth of uninfected children in their first 3 years.

OBJECTIVE: To determine the separate effects of maternal HIV infection and drug use during pregnancy on growth of uninfected children in their first 3 years. DESIGN: Retrospective analysis of measurements from health visitor records made during routine child health surveillance at 6 weeks, 10 months, and 3 years of age. Multilevel analysis allowed for between-infant variation in fitted growth lines, and adjustment for other factors. Growth was described in terms of an intercept (z score at term) and growth slopes (change in z score per year) up to, and from, 4 months. SUBJECTS: 290 case babies delivered in Edinburgh hospitals to women who reported injection drug use by either themselves or their HIV infected partner, and 186 community controls. A total of 131 (45%) of the case babies were born to women who used drugs, predominantly opiates, during pregnancy and 93 (32%) to HIV infected women. The eight infected children were excluded from analysis. MAIN OUTCOME MEASURES: Age and sex standardised z scores for height, weight, and body mass index. RESULTS: 459 (96%) of the 476 records for cases and controls were traced, yielding 1432 weight and 939 height measurements. Maternal HIV infection was not found to affect growth; at 3 years the estimated effect on weight z score was 0.16 with 95% confidence interval (-0.25 to 0.57) and for height 0.18 (-0.19 to 0.55). Drug use during pregnancy was associated with lighter babies at 40 weeks followed by depressed growth in the first four months, these infants remaining just slightly smaller at 3 years with an estimated effect on z scores of -0.5 for weight with 95% confidence interval (-0.89 to -0.11) and -0.37 (-0.72 to -0.02) for height. CONCLUSIONS: Maternal HIV infection does not adversely affect growth in uninfected infants, and the effect of drug use during pregnancy is limited to small decrease in size at 3 years.

HIV infection in women: immunological markers and the influence of pregnancy.

OBJECTIVE: To describe the influence of pregnancy on immunological marker paths and progression of HIV-infected women. DESIGN: Analysis of prospectively collected immunological and clinical data collected on 145 women reviewed at the City Hospital, Edinburgh, between 1985 and 1992 using a two-level random-effects model that allows for within- and between-patient variance. RESULTS: There were differences between the marker paths of women according to risk activity; women who had acquired HIV via injecting drug use (in addition to heterosexual intercourse) had a higher level of absolute CD4 cells, CD4% and total lymphocytes at seroconversion than those who had acquired HIV via heterosexual intercourse alone; however, immunological markers declined more steeply after seroconversion. There was no evidence that pregnancy, either before or after HIV seroconversion had an adverse effect on marker paths of HIV disease. There was a significant association between pregnancy after HIV seroconversion and post-pregnancy changes in immunological markers: an increase in the CD4% and a decrease in CD8%. However, causality cannot be implied as pregnancy itself may be associated with considerable lifestyle changes. During pregnancy the total white blood count rose due to an increase in the number of granulocytes, whereas the total lymphocyte numbers fell. The absolute CD4 lymphocyte subset counts fell progressively but the effect was due to the fall in the total lymphocyte counts, there being no influence of pregnancy on either CD4% or CD8%. CONCLUSIONS: In asymptomatic HIV infection, changes in the absolute levels of CD4 and CD8 lymphocyte counts are primarily related to changes in the other components of the white cell count because there were no changes in CD4% and CD8%. Pregnancy itself has no adverse effect on immunological markers of HIV.

HIV prevalence and risk factors in university students.

OBJECTIVE: To estimate HIV prevalence and risks in university students. DESIGN: Anonymous self-completion questionnaire and HIV survey with saliva samples. SETTING: University students at matriculation. PARTICIPANTS: All first and third year undergraduates and newly registering postgraduates at the University of Edinburgh, Scotland. MAIN OUTCOME MEASURES: HIV prevalence, sexual behaviour, condom use, drug use. RESULTS: The questionnaire responses were used to classify the 4665 respondents into four groups, ordered by risk of HIV positivity, and a sample of 2041 was selected for testing. All of the top two risk groups were tested (217 and 758 tests, respectively) as well as a random sample of the others. Five positive HIV-antibody tests were detected, all from the highest risk group. This gives an estimated rate of 1.2 per 1000 (95% confidence interval, 0.4-2.9) for all respondents. Only one of the five HIV-positives had been tested for HIV. The factors associated with HIV positivity were residence in Africa, intravenous drug use and male homosexuality. Overall, 74% of respondents reported ever having had sexual intercourse and this rate was the same for men and women. Reported intravenous drug use was very low: 0.5% for men and 0.1% for women. Condom use was more common for partners of short acquaintance, but unrelated to the number of sexual partners in the last year. CONCLUSIONS: There was no evidence of the spread of HIV infection beyond known high-risk groups in this population. This may be a result of relatively low levels of HIV risk-taking behaviour in the majority of respondents.

PIP: In Scotland during 1993-1994, 4665 first and third year undergraduates, newly registering postgraduates, and nongraduating students at the University of Edinburgh completed a questionnaire. Based on responses, the researchers categorized the students into four risk groups. One group consisted of men with male sex partners in the last year, permanent home in Africa, IV non-medically prescribed drug use, ever shared needles or works, ever paid or been paid money for sex, professionally exposed to blood. The second group include those not in the first group but had more than 3 sex partners in the last year, or persons with more than 2 sex partners in the last year and no condom use at last intercourse, or sexual intercourse with a resident of Africa. Persons who were neither in the first two groups nor the fourth group comprised the third group. Persons who never had sex and were not in group one comprised group four. They submitted saliva tests to all students in the top two risk groups and to a random sample of those in the other groups for a total of 2041 students. The researchers aimed to determine HIV prevalence and risk factors. All five HIV seropositive students were from the highest risk group for an overall HIV prevalence rate of 1.2/1000. The HIV prevalence rate for those just in the highest risk group was 22/1000. Only one of these HIV seropositive students had been tested earlier for HIV. HIV infections were limited to persons with a permanent home in Africa, IV drug use, and male homosexual intercourse. All but one HIV seropositive individual were males. 73.7% of all respondents had ever engaged in sexual intercourse. IV drug use was rare (0.5% for men and 0.1% for women). 52% of respondents used a condom during last intercourse. Condom use was associated with short acquaintance of partners. The number of sexual partners in the last year did not affect condom use. These findings indicate that HIV transmission appears to be confined to high risk groups, probably because most students did not practice risky behavior.

Reproducibility and observer variability of transrectal ultrasound measurements of prostatic volume.

Transrectal ultrasound (TRUS) is an established investigation in benign and malignant prostatic disease though the level of reproducibility of TRUS volume measurements is not known. A group of observers in the UK and the USA who were participating in linked prospective studies of benign prostatic hyperplasia each carried out measurements during real-time scanning and on a series of static TRUS images. Results demonstrated good reproducibility for measurements of antero-posterior, transverse, and longitudinal dimensions of the prostate by an experienced UK urologist; only a minor degree of interobserver variation occurred in measurements made between experienced UK and USA observers. Using static images, USA observers obtained results that were in good agreement, whereas the UK observers, only two of whom used TRUS regularly, were in poor agreement with each other. This study demonstrates a high level of reproducibility for TRUS volume measurements performed by experienced observers.