Often the research is carried out using not the most accurate of the existing methods, which allows you to get the closest to the true value of the measured value, since such reference methods (giving a reference measurement in terms of accuracy) are usually laborious, dangerous, painful or expensive. Deviations of measurement results from the true value are errors (errors) of measurement, regardless of the reason for the deviation. Random error - deviations from the true value, which on average are equal to zero, i.e. do not change the measured value. A random error makes it difficult to identify patterns, but its presence does not affect the direction and magnitude of shifts (differences, relationships) revealed in the study, of course, provided that many measurements have been made. Systematic error (SD, bias, bias,) is called the bias of the average measurement result in relation to the true value. For example, due to the peculiarities of the reagents used, different methods of measuring the concentration of individual substances in the blood (for example, troponin or glucose) give slightly different results.

Any deviation of conclusions from the truth or a process leading to such a deviation is called CO. It also denotes any bias (distortion) in the collection, analysis, interpretation, publication or review of data, leading to conclusions that are systematically different from the truth. The main mechanisms for the occurrence of displacements are as follows.

1. Systematic (one-sided) deviation of the measurement results from the true values ​​(CRM in the narrow sense, “CRM measurement”).

2. Deviation of total statistical estimates (averages, frequencies, communication measures, etc.) from their true values ​​as a result of systematic deviation of measurement results, other errors in data collection or errors in research design, data analysis.

3. Deviations of conclusions from the truth due to defects in study design, data collection, analysis or interpretation of results.

4. The tendency of procedures (in study design, data collection, analysis, interpretation, review or publication of results) to yield results or conclusions that deviate from the truth.

5. Bias, forcing, consciously or unconsciously, to choose such research procedures that lead to a deviation from the truth in a certain direction or to a one-sided interpretation of the results.

Researchers strive to obtain unbiased estimates in research, but this is not always possible, in which case the researcher must evaluate the possible amount of bias. The term SB (bias) does not necessarily imply an accusation of bias or the presence of another subjective factor, such as a desire for a particular outcome. A biased estimate can only be a consequence of the imperfection of the study design or some of its elements.

SD of a sample (sampling bias) - SD resulting from the study of a non-random sample. It should not be confused with sampling error, which is part of the overall parameter estimation error due to the random nature of the sampling. Sampling error is random, it occurs as a manifestation of the fact that each random sample from the population is different (variability of sample results, sampling variation).

The ascertainment bias is an SD associated with the inclusion in the sample of individuals or cases that do not equally represent all classes (subgroups) of the population. The reasons for this CO are varied. This may be a feature of the source where the surveyed persons come from, for example, a plant clinic (as a result, the sample will not reflect the health status of the population). It may be a way to identify people, their characteristics, in particular diagnoses, which can be influenced by customs and culture.

Selection bias is an error caused by systematic differences in characteristics between those who take part in the study and those who do not. Such an error occurs in a study that includes only volunteers (they are different from those who did not want to participate) or only hospitalized patients under the supervision of a doctor (those who died before hospitalization due to a severe course of the disease are excluded, and those who not yet ill enough to require hospitalization, and those who, due to the cost of treatment or distance, have not been hospitalized). As a result of CO selection, false connections can arise and real existing ones can be masked. SB selection is a very common problem, diverse in its manifestations. [3,4]

The response bias is the response bias caused by differences in the characteristics of those who volunteered to take part in the study and those who refused.

SD due to withdrawal from the study (bias due to withdrawals) - SD arising from the difference between the values ​​of the true and the values ​​obtained in the study, as a result of the special characteristics of the participants who left the study. For example, when studying the follow-up, it is not possible to find some of the patients. Studying the characteristics of only those who have been found can give a distorted idea even about indicators such as mortality.

SD of a series of autopsy (bias in autopsy series) - SD in the assessment of the pathological picture, resulting from the fact that the autopsy is a non-randomized sample of all deaths. For example, when analyzing the current results of postmortem autopsies, one cannot but take into account that pathologists study only half of the dead. It is possible that in the "invisible" half both the structure of the causes of death and the frequency of discrepancies with the clinical diagnosis are different.

SD of allocation of patients (allocation bias) - in experimental studies of treatment methods, an unequal distribution of patients between the compared groups is possible, as a result, for example, the frequency or speed of recovery in "mild" patients receiving a new intervention is compared with similar symptoms in more severe patients receiving standard treatment.

Publication of research results in a specific language, depending on the nature and focus of the results. There is a long-standing question as to whether there is a language bias that makes researchers choose to publish their negative results in non-English-language journals and leave their positive results to English-language journals. Some studies have shown that language limitations in systematic reviews can alter the results of the review, and in other cases, the authors did not find such bias to exist. Citation or refusal to cite research results, depending on the nature and direction of the results. Authors tend to cite positive results rather than negative or nil, and this has been established across a wide range of topics. Differential citation can lead to a perception in the community that an intervention is effective when it is not, and can lead to over-reporting of positive results in systematic reviews if those that are not cited are difficult to find. Selective aggregation of results in meta-analysis is a form of citation bias that is particularly insidious in its potential to influence knowledge. To minimize bias, pooling results from similar but distinct studies requires a careful search of all relevant studies. That is, a meta-analysis (or pooling of data from multiple studies) should always come from a systematic review (rather than a selective review of the literature), even if the systematic review is not always associated with meta-analysis.

Regards, Sergey

Sara Hamidi

publication bias

Publication and Selection bias are always potential concern specially for meta-analyses.

With the proliferation of meta-analyses in the medical literature have come conflicting studies. In addition, observance of guidelines for the performance of meta-analyses has been spotty. Bias may explain conflicting studies and differentiate carefully performed meta-analyses from others. Meta-analysts may fail to anticipate biases which threaten their study's validity. The three stages at which bias can be injected into a meta-analysis are finding studies, selection of the identified studies for the meta-analysis and extraction of data from the selected studies. https://pubmed.ncbi.nlm.nih.gov/1624971/

Kindly check https://warwick.ac.uk/fac/cross_fac/bridges/scientificreproducibility/warwick_seminar1.pdf