Reporting of research is done to apprise others of the new development. This objective is more effectively achieved when the communi-cation contains enough details of the methodology and all other aspects so that the reader is convinced about the validity of the results, can assess their generalizability, and is able to replicate the results if needed.

Statistical methods have become an essential component of all empirical research publications, more so for biomedical research that confronts enormous uncertainties due to biological and environmental variability, sampling fluctuations, epistemic bottlenecks, and biases. Science requires that these methods are fully reported with complete accuracy so that the results could be fully appraised for validity, reliability and generalizability, and evidence-based medicine is strengthened. To meet this objective, Statistical Analyses and Methods in Published Literature (SAMPL) guidelines [1] have been prepared for statistical reporting in biomedical publications. However, these guidelines have some lacunae. For example, these guidelines mentioned about identifying the variables separately for Primary Analysis, for Reporting Hypothesis Tests, for Reporting Association Analysis, for Reporting Regression Analysis, and several others. Some of the essentials such as comparability of groups and robustness have been missed. These guidelines need to be reorganized on the lines of other reporting guidelines such as CONSORT. This communication proposes substantial improvement of these guidelines to make them more organized, compact, and easier to adopt.

Performing the appropriate analysis is different from accurately describing it, and there is no way for a third person to assess what was actually adopted except by reporting in the publications. It is expected that much of these errors can be avoided by improved statistical reporting.

Statistical reporting in biomedical publication is an important part of the Material and Methods section but it also affects the way the results are understood and interpreted. Several studies have observed that the statistical reporting in some biomedical publications is inadequate [11-14]. These studies suggest that this inadequacy generally occurs at three levels: (i) incomplete reporting leaving out room for readers to impute guess: (ii) willful or inadvertent erroneous reporting that has potential to arouse suspicion about the results; (iii) and inadequate interpretation of the statistical results. Much of this deficiency can be effectively addressed if the publications adhere to a standard guideline of items for reporting of the statistical methodology so that it is fully reported in a proper manner without missing any essential component. This may also encourage researchers to use the right statistical methods at various stages of their research.

Much of the clarity in reporting comes from clear statements about how the data were collected; what analysis was done how; why that particular analysis was appropriate for the problem in hand; and how the conclusion was drawn. Statistical methods in an empirical research can be intricate multivariate and multilevel analyses or can be specialized such as time series analysis whose description is admittedly challenging, but many errors have been observed in basic methods used in biomedical publications [21]. As these are basic methods, there is a tendency to use and describe them without sufficient care [22]. The proposed guidelines are focused on these basic methods only.

In view of the common occurrence of statistical errors in biomedical publications, attempts have been made in the past to present guidelines for statistical reporting [17,23,24]. Subsequently, Lang and Altman [1] compiled a set of guidelines for basic statistical reporting for articles published in biomedical journals. They called it "Statistical Analyses and Methods in the Published Literature" or the SAMPL Guidelines, and these are now part of the EQUATOR network [9]. The authors acknowledge that these guidelines are limited to the basic methods but consider them sufficient to prevent most of the reporting deficiencies as the basic methods are also the most commonly used methods. The first guiding principle for these guidelines is that the statistical methods should be described with sufficient detail for a knowledgeable reader to verify the reported results if the data are provided to him, and the second principle is to report enough details of the descriptive statistics from which other indicators such as relative risk and odds ratio are derived.

Besides that the current SAMPL guidelines have not included some of the basic methods such as comparability of groups and robustness of results,these are also repititive. They also need to be reorganized in a compact form just as are other statements such as CONSORT, STROBE and STARD. These statements have been revised from time to time as and when new knowledge is acquired and it is time to revise the SAMPL guidelines as well to make them more organized, compact and easy to adopt. We have undertaken this exercise and the guidelines have been substantially revised in content and format (Table I). Most notable change is the complete reorganization of format to a numbered list for easy adoption. This also removes much of duplication. Other notable changes are inclusion of background information of the subjects, reporting of standardized rates (where needed) for comparability, robustness of results, not reporting mean and SD for extremely small sample size, and careful reporting of cause-effect inference. There are several other changes to make the guidelines more comprehensive and easy to understand. The reorganization is in terms of a list with 16 items, many with sub-items, which can also be used as a checklist. First 13 items will be required by almost any biomedical publication based on empirical data and the remaining 3 items are for specialized methods. To avoid duplication, there is no separate item on ANOVA and ANCOVA as reporting of these is included in other items. Bayesian analysis is also excluded as it is not a commonly used method in biomedical publications. Hazard ratio is excluded because of its specialized nature. Now, there is a clear demarcation of items to be reported for each analysis undertaken by the researcher although we continue to adhere to the principles enunciated earlier [1].

TABLE I Improved SAMPL Guidelines for Reporting Basic Statistical Methods in Biomedical Publications 

This revision is also restricted to the reporting of the basic methods. The advanced methods such as Cox regression, cluster analysis, and multivariate analysis of variance (MANOVA) are excluded in the hope that a qualified biostatistician will be involved when such advanced methods are used and the reporting will be adequate. The basic methods covered by these guidelines are generally used by those also who use advanced methods. To retain the focus, other methodological aspects such as design, allocation and randomization as well as issues relating to proper graphs, diagrams and tables are excluded. These suggested guidelines continue to be described in a manner that a statistically literate medical researcher can adopt without much help of a statistician. As in the case of original version [1], this suggested revision too is not prepared by a ‘formal consensus-building process’ but is prepared after consulting various other guidelines [24-27].

We hope that the editors of the biomedical journals will incorporate these guidelines in their instructions so that the reporting of basic statistical methods can improve and evidence-based results are reported. The real solution to poor statistical reporting will come when authors and statisticians learn more about research methodology and appropriate analysis, and also learn to communicate it properly [11]. Deficient statistical reporting underscores the need to expose the medical researchers to detailed texts [28,29] and structured biostatistics courses so that the methodology and reporting can improve.

Funding: None; Competing interest: None stated.