Design and data source

We conducted a methodological study encompassing all original analytical articles published in Peruvian medical journals indexed in SciELO-Peru, Scopus, or Medline, between 2021 and 2022.

We included all medical journals that were indexed in SciELO-Peru, Scopus, or Medline during data extraction (April 2023). These were nine: “Acta Médica Peruana”, “Revista Médica Herediana”, “Revista de la Facultad de Medicina Humana”, “Horizonte médico”, “Revista del Cuerpo Médico del Hospital Nacional Almanzor Aguinaga Asenjo”, “Revista de Neuro-Psiquiatría”, “Revista Peruana de Ginecología y Obstetricia”, “Revista de Gastroenterología del Perú”, and “Revista Peruana de Medicina Experimental y Salud Pública”.

Subsequently, we examined the official websites of these journals to scrutinize every article published in their volumes corresponding to the years 2021 and 2022. Our inclusion criteria encompassed original studies on human subjects, incorporating analytical assessments (i.e., associations between variables using any bivariate association analyses). We excluded secondary studies such as systematic reviews or scoping reviews, as well as validation studies.

Data collection process and variables

Two authors (NNA and DGL) independently screened the articles carefully to identify those who met the inclusion criteria. In cases of disagreement, a third author was required to resolve any discrepancy (DFG or ATR). Data extraction was performed into an Excel spreadsheet and was independently carried out by two of the authors (NNA and DGL).

The general variables extracted from each article included: journal name, year of publication, study design or field of study. This field was categorized into clinical or public health types. Clinical articles focus on diagnosing, treating, and managing patients individually or in groups. Public health articles address disease prevention, health promotion, policy, epidemiology, and more. Additional variables encompassed time between submission and acceptance of the manuscript, number of authors, and presence of any Peruvian authors.

The statistical variables assessed were: statistical software, sample size, and employed statistical methods. These include measures of effect and controlling for confounders. Measures of effect included linear regression beta coefficient [β], odds ratio [OR], prevalence ratio [PR], relative risk [RR], hazard ratio [HR]. Methods employed for cofounder control can be statistical or epidemiological approaches. Statistical methods rely on p-values or other statistical methods, while epidemiological approaches involve theoretical rationale and clinical relevance. These variables were chosen according to previous studies [4, 5, 8–10]. This data was gathered by thoroughly reviewing each article, with particular focus on the methods and results sections, although additional sections were consulted when required.

Statistical analysis

Data entry was conducted using Microsoft Excel, and subsequent statistical analyses and graphics were carried out using R software, version 4.1.0 (R Foundation for Statistical Computing, Vienna, Austria). To provide a comprehensive overview of the data, descriptive statistics were employed. The results were presented in the form of either medians and interquartile ranges or means and standard deviations, depending on the normality distribution, which was assessed using the Shapiro-Wilk test. Categorical variables were summarized using counts and percentages. Bar graphs were generated to illustrate the frequency of the softwares, statistics, graphics and the calculation of effect measures in the included articles.

Sample size and statistical power

Whether the sample is collected for the study or obtained from a pre-existing database, determining the sample size is crucial. This ensures to obtain the minimum number of subjects necessary to detect specific associations. Similarly, in secondary data analysis, assessing whether non-significant findings are due to a lack of statistical power is important.

However, a significant proportion of original articles (78%) show deficiencies in specifying the calculation of sample size or statistical power. Sample size addressing may omit essential computation details, hindering replicability despite its acknowledgment in the analysis. Kumar et al. reported similar results examining statistics application trends in 300 biomedical research articles. They found that, despite the mention of sample size in 95% of the articles, reproducibility was limited to only 7%. In the remaining articles, neither the precise mention of the sample size estimation technique nor an adequate description was provided [11]. Likewise, in 448 Balkan Medical Journal studies, sample size reporting was 8.1% in cross-sectional studies, 13.8% in cohorts, and 16% in clinical trials [12]. Even in high-impact general medical journals, only one-third of articles adequately describe sample size calculations [13].

Measures of effect

Our study found that 47% of articles lacked measures of effect. A parallel trend was identified in a study that assessed 60 observational studies in high-impact occupational medicine and health journals [14]. Although all studies reported outcome events or summary measures, only 60% provided unadjusted estimates, and merely 43.3% included confounder-adjusted estimates [14]. Similarly, an examination of JAMA articles between 1990 and 2010 uncovered an increase in incorporating epidemiologic statistics, reaching almost 30% [3]. This highlights a gap in reporting standards, emphasizing the need for improved consistency in presenting both unadjusted and adjusted estimates. In contrast, an analysis of 113 studies from a Colombian journal (Biomedica, 2000–2017) revealed nearly 90% reported measures of association coefficients [4].

Measuring the strength of risk factor-outcome associations is an important goal in epidemiological and clinical research [15]. Effect sizes—both unstandardized and standardized from individual studies—should be reported for inclusion in systematic reviews and meta-analyses [16]. However, our analysis revealed that only 13.4% of articles provided criteria for selecting confounding variables, essential to adjusted analyses. These can be chosen based on established causal relationships, statistical associations (p-value < 0.05), or an epidemiological approach. In our study, confounding variables were primarily selected through the last two methods.

Adequate management of confounding variables is essential for isolating the true impact of the exposure under investigation (causal inference) [17–20]. Specifying criteria for selecting confounding variables in an adjusted model analysis ensures systematic identification and control, enhancing research reliability. Unclear descriptions of statistical techniques addressing confounding factors and biases hinder reader assessment of result validity.

Despite explicit reporting recommendations for handling missing data within the STrengthening the Reporting of OBservational studies in Epidemiology (STROBE) guidelines [21], we found no articles adhering to these recommendations. Failing to handle missing data appropriately may lead to incorrect conclusions and potential harm to patients if clinical decisions are based on incomplete or biased information [16].

Even among high-impact medical journals, deficiencies persist in reporting STROBE items. For instance, high-impact occupational medicine and health journals reported all STROBE items in only 63% of articles, while otorhinolaryngology journals adequately reported a mean of 51.4% of STROBE items [14, 22]. Our findings underscore the need for enhanced reproducibility and quality reporting in observational studies published in Peruvian journals. Therefore, addressing this issue should be a central point for both authors and editors when reporting or reviewing observational studies.

In scientific research, there are three main approaches: describing, explaining and predicting [23]. Describing refers to characterizing and summarizing data without attempting to establish causal relationships, focusing on the who, what, when and where of the phenomena studied. Explaining goes beyond description, seeking to understand the why and how of the phenomena, which implies identifying causal relationships and underlying mechanisms. To this end, it is crucial to adequately control for confounding variables and biases. Prediction, on the other hand, focuses on the use of data and models to anticipate future events or behaviors, based on patterns identified in historical data. This approach requires robust models that are able to generalize to new data and situations.

Despite the extensive use of statistical analysis in the studies reviewed, no research was found that adequately assessed causality. Many of the articles were limited to describing associations using bivariate analyses and regression models, without comprehensively addressing the criteria necessary to infer causal relationships. This includes control for selection bias and adequate control for confounding variables, as well as insufficient explanation of the methods used for this selection [24]. This limitation suggests that, although different statistical techniques were applied, they were not used with the necessary rigor to establish causality, which underlines that many of the studies have a descriptive approach.

Inadequate presentation of the data

Our analysis identified a frequent oversight in the presentation of data within the examined articles. The mention of the statistical software was absent in 9.9% of the articles. Furthermore, 48.6% of the articles presented tables/figures that included p-values but in which the statistical tests used were not specified.

Tables play a crucial role in presenting research results. Even when presented independently, they should include essential elements necessary for communicating information, such as population parameters, alongside confidence intervals, p-values, and the utilized test statistics [25]. Inclusion of these components in tables facilitates interpretation of results for readers. The presentation of p-values and confidence intervals enables clinicians to discern the statistical significance and understand the differences in outcomes between two groups [26]. While p-values indicate statistical significance, confidence intervals establish the threshold for the magnitude of the effect. Both of these statistical measures are integral in the interpretation of clinical outcomes. Therefore, ensuring their inclusion in tables enhances the clarity and meaningfulness of research findings.

Peruvian journals

Peru encounters significant challenges regarding the representation of its journals in key international databases like Scopus and Medline. At the time of the study, only two Peruvian journals, "Revista Peruana de Medicina Experimental y Salud Pública" and "Revista de Gastroenterología del Perú," are indexed in both Medline and Scopus. This aligns with the current national situation, where we have approximately 92 health science journals, with around 95% remaining unindexed [27].

This scenario is significantly impacted by crucial factors, stemming from a shortage of researchers and a prevalent inclination to publish their research in outside journals indexed in international databases such as Scopus or Medline [28]. Additional challenges include insufficient financial support and promotion for national medical journals. Collectively, these factors diminish visibility and recognition both regionally and internationally.

Our study highlights significant concerns regarding the methodological reporting of statistical processes in articles published in Peruvian journals. Introducing a statistical review of the submitted articles as an integral part of the editorial process, as suggested in our study, could be a replicable strategy for enhancing the reporting quality of research articles in diverse international contexts and improving confidence in their results [29].

National journals are important to spread studies of local interest, which may be of special interest for local practitioners and stakeholders [30]. Nevertheless, as revealed by our study, studies published in local Peruvian medical journals need to improve their statistical report, which ultimately may underpin their credibility and usage.

Efforts to identify and improve the statistical deficiencies of Peruvian medical journals can foster a more inclusive and comprehensive representation of diverse perspectives in the global scientific community. By recognizing the challenges faced by Peruvian journals and proposing solutions, our study aims to contribute to a broader conversation on strengthening the integrity and impact of research from developing regions, ultimately fostering a more interconnected and equitable landscape for scientific knowledge dissemination.

Limitations and strengths

The findings of this study shed light on critical aspects of statistical reporting in Peruvian biomedical articles. However, some limitations should be taken into account when interpreting the results. Firstly, the temporal scope of the study, which focused on publications from 2021 and 2022 to reflect the current state of the topic, but may not be generalizable to different time periods, as statistical reporting may change over an extended time. Additionally, while the study reviewed statistical reporting, it did not thoroughly cover all aspects of research methodology.

It is essential to recognize that despite the limitations reported in the medical Peruvian journals, they do not fully represent the medical research produced in Peru, since the majority of Peruvian biomedical articles are published in international journals, which were not the focus of this investigation [6, 31]. Analyses of Peruvian research revealed that only 11.3% of COVID articles up to 2021 [31] and just over 19.1% of public health articles up to 2020 [6] were published in Peruvian journals. However, it is relevant to note that the primary objective of our study was to assess Peruvian journals to provide insights for their improvement.

Turning to the strengths of the study, to our knowledge, this is the first study conducted in Peru and within Spanish journals assessing statistical reporting in medical literature. Furthermore, a thorough article analysis was carried out in duplicate to ensure optimal data extraction.