Summary of the article
In rheumatoid arthritis, the allocation of progression scores is severely skewed, necessitating complex statistical analysis approaches, with different methodologies yielding different results.
Hypothesis: Most individuals with early rheumatoid arthritis who engage in clinical experiments have modest damage scores and little or no progression.
The Methods: Three databases were reviewed for clinical experiments and observational research involving rheumatoid arthritis that detailed radiographic analysis methodologies and compared at least two groups (Mahmood et al., 2019). The frequency of joints examined, the amount of detail and the probability of healing varied between methods. Larsen and its variants are typically cited as the most practicable, and Sharp van der Heijde score as the most dependable and change-sensitive.
The Findings: Among the 5980 publications discovered, 225 were qualified for data extraction. Thirty-nine papers employed metric approaches such as t-tests and linear regression, with 18 percent accounting for the uneven distribution. Continuous data was classified and evaluated in 53 investigations using ordinal or binomial techniques, logistic regression, and chi-square tests. Two researchers used a Poisson distribution to treat the result as a count event variable.
In conclusion, the analytic technique of radiographic progression in contemporary rheumatoid arthritis clinical experiments and observational research is very heterogeneous. The majority of these studies employ primary, inadequate, or unsuitable methodologies.
Search Strategy
The current article used the suggested search approach for narrative reviews. As a result, the PubMed, Web of Science, and Cochrane databases were examined using the terms “statistical analysis,” “errors,” “rheumatoid arthritis,” and “radiological data,” along with relevant logical connections. The narrative review comprised articles produced in the English language. Correspondence to editors, remarks, summaries, and unpublished works was among the first to be excluded (Coskun, 2019). After the literature research, the received papers were evaluated for suitability, and duplicate or irrelevant elements were eliminated. The selected articles’ references were examined, and relevant articles were included in the narrative review.
Rheumatoid arthritis is a severe systemic inflammatory condition characterized by persistent symmetric polyarthritis of the feet and hands. Rheumatoid arthritis mostly affects the body immune system of a body. The information is accessible in three databases: Pubmed, Embase, and Cochrane. The comprehensive literature search yielded 8727 titles. Thomson Endnote X7’s automated function discovered and eliminated 2747 duplicates (Schwen & Rueschenbaum, 2018). A sum of 5980 publications was reviewed by abstract and title, with 308 examples for full-text analysis. Fifty-five were eliminated for various reasons, including a lack of statistical comparability, descriptive research, and language usage.
Nevertheless, interpretation may be problematic if log-transformation results in a bell-shaped normally distributed. Because log transformation just marginally enhances the distribution, a few researchers believe that parametric longitudinal analysis and regression analysis techniques are still more rigorous for this issue. Given a sufficient sample size, the distribution of a sample of a skewed distribution will be significantly more regular than the central limit theorem (Carmona-Bayonas et al., 2018). The skewness reduces the efficacy of parametric tests since the mean is no longer a valid estimate of the normal distribution because absolute values heavily influence it.
Non-parametric tests, on the other hand, function well with skewed data distribution and ranges that are better reflected by the average. Based on available data, the assumption is that neglecting the nature of the distribution could enable one to proclaim a difference depending on the mean, even when such parameters are insufficient or even skewed representations of the data distribution. It then leads to skewed interpretations (Bahar et al., 2018). The data quality is poor, with most research ignoring the usual range of radiological data.
Having zero-inflated data means that the data is positively skewed. The variability begs whether multiple research findings can be compared to obtain superior outcomes. Because the datasets trail off more steeply on one edge than the other, a skewed distribution is hardly normal or symmetric. Skewness in data sets is commonly used in business to describe sizes using positive values such as sales or assets. It is because datasets cannot be lower than zero, setting a barrier on one side; however, there is no distinct upper border. Consequently, multiple data values are clustered at zero, and they get progressively less and fewer as one proceeds to the right side of the histogram.
Bio-statistical Study Design
The research hypothesis should be evaluated on a representative sample and with the proper study design. In terms of technique, the research design is analytical and observational. The link across variables is quantified in analytical investigations (Bahar et al., 2018). This study’s analytical analyses were meant as observational and hence non-experimental. Analytic observational researchers study the consequences of every exposure without the researchers actively participating.
Errors are unavoidable in science and associated research activities. Statistical precision is required to overcome inaccurate research and reporting techniques. Rheumatology is a rapidly expanding field of medicine with significant research possibilities. Research is required to determine the unknown themes in rheumatology (Schwen & Rueschenbaum, 2018). These research efforts’ outcomes must be published to disseminate scientific information to other researchers, healthcare practitioners, and patients.
Nevertheless, publications’ accuracy, dependability, and overall performance are critical. Statistical precision is essential for conducting dependable and high-quality research. As a result, it is critical to ensure statistical relevance from the start of research until the completion of the entire procedure.
Each study is built on a research question. Following that, a hypothesis is proposed in response to this query, as the hypothetico-deductive technique is used in most medical research initiatives. A hypothesis is a verifiable assertion that the researchers develop. It is the theoretical working tool that connects theory to experience. Hypotheses are structural statements that provide a relational assertion and a speculative explanation of occurrences to the researcher.
We argue that there are better approaches to handle a skewed distribution, although our review shows that they are rarely adopted. Researchers use Poisson regression to summarize two studies where radiological data is considered a counting factor (Turkiewicz et al., 2018). A Poisson distribution has a right-skewed distribution. Polar alignment in binomial regression can be employed when there is excessive volatility in the count data, which typically happens whenever there is a surplus of zeros. Nevertheless, when there is an enormous surplus of zeros, all approaches provide a zero-inflated alternative.
Statistical Findings
Twenty-eight papers presented multi-study analyses, whereas 225 reports detailed single research. The latter included 160 (71 percent) clinical attempts; the rest were observational studies. Over two groups were compared in 43 percent of the individual trials. Most studies (n = 110, 49 percent) had sample sizes ranging from 100 to 500 people. Twenty-seven publications (12%) had a size sample of under 100 people, while 85 (38%) had a representative sample of over 500 people. Foot radiographs were included in almost all research (97%). 39 (17%) of the 225 articles on single investigations employed parametric approaches to evaluate the data, like t-tests, ANOVA, or linear regression (Turkiewicz et al., 2018). 35 out of the 39 research studies had a sample size of over 100, while 13 had a representative sample greater than 500.
The Kruskal-Wallis test is an extension of the Mann-Whitney U test that compares non-normal variables in several groups. The Kruskal-Wallis test is the parametric version of the Kruskal-Wallis test regression analysis. If the response variable is the ultimate determinant, then the discrepancies between the measured data and the projected answers using regression must be normally distributed. As a result, the dispersion of the predictor variables is irrelevant. There are various sophisticated strategies for dealing with similar problems, like Poisson regression. However, they necessitate particular assumptions, and it is frequently preferable to consider the result variable as continual.
The initial systematic literature review concerning analyzing radiological data of rheumatoid arthritis research demonstrates heterogeneity in data analysis techniques under radiographic progression. Most research studies ignore the distribution of radiological data. The heterogeneity presents the inquiry into the outcomes of various studies that can be compared (Brown et al., 2018). Parametric tests help compare two matched clusters, and far more than two matched clusters are matched t-tests and replicated metrics analysis of variability. Friedman’s tests and Wilcoxon’s approved rank test are nonparametric equivalents of these assessments. The independent t-test samples and ANOVA are parametric assessments that compare two unattached groups and over two unparalleled groups. Nonparametric variants of these tests include the Kruskal-Wallis test and the Mann-Whitney U test.
Recommendations
The research study had several limitations, like limiting the search to the past ten years in the study of contemporary practice. As such, it is not given that the study captures all applied procedures. However, older studies are unlikely to have used more useful analytical methods. As a result, future studies should expand the period. Secondly, the research study cannot incorporate conference abstracts since the most critical aspects of analytical procedures would probably not be available. Three articles written in a language other than English were not included for worthwhile causes, thereby missing some information that could be essential in the study.
Every statistical analysis paragraph should be exceptional because one generic piece would not handle all research studies. About ten percent of the articles do not give statistical methodology. This part of the research paper should include data analysis techniques, statistical software packages used, and the accepted significance level (Karadeniz et al. 2019). When using several statistical techniques, the researchers should specify which test is helpful for what data rather than using phrases like where appropriate. When performing several comparisons, numerous comparison-level corrections need to be given.
In simulations with a fluctuation in the lower threshold of the scale disparity in the zeros, Tobit analysis is advised. A negative binomial or Poisson approach can be employed when the counts are skewed towards the right. Zero-bloated models work well for counts with an overabundance of zeros. These models are only applicable to cross-sectional datasets. The two-part approach can be employed in longitudinal investigations with an overabundance of real zeros.
My view as the author
RA is a chronic condition that needs treatment to slow disease progression. The early symptoms are connected to joint inflammation, but long-term consequences might include extra-skeletal signs. I will include graphs and pie charts because researchers are concerned about presenting satisfactory statistical results. Text, tables, and figures are all examples of data display methods. Essentially, researchers start publishing the study results as a text within the summary of the research work. At that time, being succinct and informative was crucial (Turkiewicz et al., 2018).
First, a brief demonstration of the sampling results should be provided in graphs and pie charts. The author in this research did not utilize graphs or pie charts for better understanding. The outcomes with relation to secondary targets may follow the primary discovery. I will include findings and applicable p-values, predictive relevance, and confidence intervals within the executive summary whenever possible. There are no statistical errors since this is continuous research open to modifications.
Data availability
Twenty-eight papers presented multi-study analyses, whereas 225 reports detailed single research. The latter included 160 (71%) clinical studies; the rest were observational studies. Over two groups were compared in 43% of the standalone trials. Most studies (n = 110, 49%) had large samples ranging from 100 to 500 people. Twenty-seven publications (12%) used a sample group of fewer than 100 people, while 85 (38%) had a representative sample of over 500 people (Brown et al., 2018). Foot radiographs were included in all studies that, represent 97%. Continuous data were classified and analyzed using ordinal or binomial techniques, such as chi-square testing and regression analysis, in 53 (24%) of the research.
The accurate data and statistics were applied, and radiographs were viewed randomly in 99 research studies (44%). Two readers graded the radiographs in most trials (134, 60%). In seven of them, one reader viewed all radiographs, while the second took a sample (Mahmood et al., 2019). The majority of the remaining investigations used a single reader. There was no data on the number of viewers or their reading sequence in 80 submissions (36%).
I want to gather data from previous centuries since, thanks to discoveries in new medications in the late twentieth and early twentieth centuries, results for people with rheumatoid arthritis improved considerably. As research outcomes in clinical studies of novel medicines, patient-reported outcome measures increasingly play a crucial role in drug discovery, leading to greater attention to the patient’s viewpoint, medication safety, and treatment results in clinical practice (Gosselin, 2019). Many planned cohorts for rheumatoid arthritis patients and databases of biological anti-rheumatic medications that were disease-modifying have been extensively performed in the United States, Asia, and Europe in response to these demands.
The examination of studies used existing data to discover potentially substantial heterogeneity and insufficiency of analytic methodologies, which treatment enhancements may address. A consistent statistical technique is preferred to derive an accurate conclusion from the present information and compare findings across investigations. Although it was not the primary focus of this investigation, the scoring technique was critical. The included studies used uniform grading procedures and assessed the feet virtually. The study employed statistical approaches that differed in the various joints and analyzed the amount of detail. It also used the approach and the probability of healing to validate the author’s conclusion that the radiography progression modeling approach used in contemporary rheumatoid arthritis clinical studies and observational research is poor and varied. Furthermore, critical information is poorly recorded.
References
Bahar, B., Pambuccian, S. E., Barkan, G. A., & Akdaş, Y. (2019). The use and misuse of statistical methods in cytopathology studies: review of 6 journals. Laboratory Medicine, 50(1), 8-15.
Brown, A. W., Kaiser, K. A., & Allison, D. B. (2018). Issues with data and analyses: Errors, underlying themes, and potential solutions. Proceedings of the National Academy of Sciences, 115(11), 2563–2570.
Carmona-Bayonas, A., Jimenez-Fonseca, P., Fernández-Somoano, A., Álvarez-Manceñido, F., Castañón, E., Custodio, A.,… & Valiente, L. P. (2018). Top ten errors of statistical analysis in observational studies for cancer research. Clinical and Translational Oncology, 20(8), 954-965.
Coskun, I. (2019). Reliability of real-world data. Rheumatology International, 39(3), 583–584.
Gosselin, R. D. (2019). Guidelines on statistics for researchers using laboratory animals: the essentials. Laboratory animals, 53(1), 28-42.
Karadeniz, P. G., Uzabacı, E., Kuyuk, S. A., Kesin, F. K., Can, F. E., Seçil, M., & Ercan, İ. (2019). Statistical errors in articles published in radiology journals. Diagnostic and Interventional Radiology, 25(2), 102.
Mahmood, S., van Tuyl, L., Schoonmade, L., Landewé, R., van der Heijde, D., Twisk, J., Boers, M. (2019). A systematic review of rheumatoid arthritis clinical studies: Suboptimal statistical analysis of radiological data. Semin Arthritis Rheum, 49(2):218-221.
Schwen, L. O., & Rueschenbaum, S. (2018). Ten quick tips for getting the most scientific value out of numerical data. PLoS computational biology, 14(10), e1006141.
Turkiewicz, A., Luta, G., Hughes, H. V., & Ranstam, J. (2018). Statistical mistakes and how to avoid them–lessons learned from the reproducibility crisis. Osteoarthritis and Cartilage, 26(11), 1409-1411.