Subgroup analysis and other (mis)uses of baseline data in clinical trials

BACKGROUND

Baseline data collected on each patient at randomisation in controlled clinical trials can be used to describe the population of patients, to assess comparability of treatment groups, to achieve balanced randomisation, to adjust treatment comparisons for prognostic factors, and to undertake subgroup analyses. We assessed the extent and quality of such practices in major clinical trial reports.

METHODS

A sample of 50 consecutive clinical-trial reports was obtained from four major medical journals during July to September, 1997. We tabulated the detailed information on uses of baseline data by use of a standard form.

FINDINGS

Most trials presented baseline comparability in a table. These tables were often unduly large, and about half the trials inappropriately used significance tests for baseline comparison. Methods of randomisation, including possible stratification, were often poorly described. There was little consistency over whether to use covariate adjustment and the criteria for selecting baseline factors for which to adjust were often unclear. Most trials emphasised the simple unadjusted results and covariate adjustment usually made negligible difference. Two-thirds of the reports presented subgroup findings, but mostly without appropriate statistical tests for interaction. Many reports put too much emphasis on subgroup analyses that commonly lacked statistical power.

INTERPRETATION

Clinical trials need a predefined statistical analysis plan for uses of baseline data, especially covariate-adjusted analyses and subgroup analyses. Investigators and journals need to adopt improved standards of statistical reporting, and exercise caution when drawing conclusions from subgroup findings.

Patient assertiveness and physician decision-making among older breast cancer patients

The objective of this study was to determine whether assertive patient behavior influences physician decision-making in the treatment of older breast cancer patients. One hundred and twenty-eight physicians saw videotapes depicting women seeking care for breast cancer and then recommended evaluation and treatment plans. Identical scripts were used, but the age, race, socioeconomic status, mobility, general health, and assertive behavior of the patients were experimentally varied along with the physician's specialty and length of practice. No direct effects of assertive patient behavior were seen. However, black, comorbid, and lower SES women were more likely to have full staging of their tumors ordered when they made an assertive request. Treatment recommendations also showed an interaction of assertiveness with patient's age and social class as well as physicians' specialty. The results indicate that a moderately assertive patient request may change provider behavior, although the effects of assertiveness vary most by what type of patient demonstrates this behavior. In particular, assertiveness led to more careful diagnostic testing for patients who came from groups that are "disadvantaged."

Intensity and correlates of fear of falling and hurting oneself in the next year: baseline findings from a Roybal Center fear of falling intervention

Fear of falling is highly prevalent and associated with restricted activity. To help inform design of interventions, the authors examined the correlates of this fear. Data came from baseline information on subjects in a community-based falls intervention study (N = 392). In a multivariate model, lower levels of fear of falling and hurting oneself in the next year were related to being younger, having higher levels of dysfunction, and having lower levels of perceived ability to manage falls, with the last two remaining significant even after controlling for generalized fearfulness. When analyzing specific domains of dysfunction, higher levels of fear of falling and hurting oneself in the next year were associated with higher levels of physical dysfunction. The findings lend support to the expanding awareness that fear of falling may imperil quality of life and suggest the importance of interventions aimed at enhancing ability to manage falls.

Physician variability and uncertainty in the management of breast cancer. Results from a factorial experiment

OBJECTIVES

The purpose of this research was to determine the influence of patient and physician characteristics on physicians' level of variability and certainty in breast cancer care.

METHODS

One hundred twenty-eight physicians viewed a videotape of a simulated physician-patient interaction in which the patient has an "atypical" breast lump. Six patient characteristics (age, race, socioeconomic status, physical mobility, comorbidity, presentation style) were manipulated experimentally, resulting in a balanced set of 32 different "patients." Physician subjects were recruited to fill four equal strata defined by specialty (surgeons versus nonsurgeons) and experience (< or = 15 or > 15 years since graduation from medical school).

RESULTS

More than half of the physicians offered a diagnosis of benign breast disease, a third offered a diagnosis of breast cancer, and the rest believed that the patient had a normal breast or something "other." Results also indicated that physicians' level of certainty and test ordering behavior varied with the diagnosis that was offered. Of the six patient characteristics, only socioeconomic status influenced physician certainty; physicians were more certain of their diagnosis when the patient was of a higher socioeconomic status. Surgeons were found to be more certain of their diagnosis compared with nonsurgeons. However, surgeons were less likely to order radiologic tests or a tissue sample for metastatic evaluation than were nonsurgeons.

CONCLUSIONS

Overall, physicians displayed considerable variability and uncertainty when diagnosing and managing possible breast cancer.

Nonmedical influences on medical decision making: an experimental technique using videotapes, factorial design, and survey sampling

OBJECTIVE

To study nonmedical influences on the doctor-patient interaction. A technique using simulated patients and "real" doctors is described.

DATA SOURCES

A random sample of physicians, stratified on such characteristics as demographics, specialty, or experience, and selected from commercial and professional listings.

STUDY DESIGN

A medical appointment is depicted on videotape by professional actors. The patient's presenting complaint (e.g., chest pain) allows a range of valid interpretation. Several alternative versions are taped, featuring the same script with patient-actors of different age, sex, race, or other characteristics. Fractional factorial design is used to select a balanced subset of patient characteristics, reducing costs without biasing the outcome.

DATA COLLECTION

Each physician is shown one version of the videotape appointment and is asked to describe how he or she would diagnose or treat such a patient.

PRINCIPAL FINDINGS

Two studies using this technique have been completed to date, one involving chest pain and dyspnea and the other involving breast cancer. The factorial design provided sufficient power, despite limited sample size, to demonstrate with statistical significance various influences of the experimental and stratification variables, including the patient's gender and age and the physician's experience. Persistent recruitment produced a high response rate, minimizing selection bias and enhancing validity.

CONCLUSION

These techniques permit us to determine, with a degree of control unattainable in observational studies, whether medical decisions as described by actual physicians and drawn from a demographic or professional group of interest, are influenced by a prescribed set of nonmedical factors.

Patient, physician and presentational influences on clinical decision making for breast cancer: results from a factorial experiment

This study examines the influence of six patient characteristics (age, race, socioeconomic status, comorbidities, mobility and presentational style) and two physician characteristics (medical specialty and years of clinical experience) on physicians' clinical decision making behaviour in the evaluation treatment of an unknown and known breast cancer. Physicians' variability and certainty associated with diagnostic and treatment behaviour were also examined. Separate analyses explored the influence of these non-medical factors on physicians' cognitive processes. Using a fractional factorial design, 128 practising physicians were shown two videotaped scenarios and asked about possible diagnoses and medical recommendations. Results showed that physicians displayed considerable variability in response to several patient-based factors. Physician characteristics also emerged as important predictors of clinical behaviour, thus confirming the complexity of the medical decision-making process.