Family medicine residents’ knowledge and attitudes towards biostatistics, Taif, Kingdom of Saudi Arabia

ABSTRACT

Background:

It is essential for practicing clinicians to have core knowledge of biostatistics. However, surveys indicated that clinicians’ attitudes towards biostatistics are negative. Despite its importance, little is known about the knowledge of and attitudes towards statistics among trainees in family medicine, particularly in Saudi Arabia. The current investigation attempts to evaluate knowledge and attitudes held by family medicine trainees in Taif and explore their correlates.

Materials and Methods:

This was a descriptive, questionnaire-based, cross-sectional study of residents in family medicine training programme in Taif, Saudi Arabia. We used Poisson regression modelling to evaluate the effect of background factors on knowledge and attitudes towards biostatistics.

Results:

The study included 113 family medicine trainees at different levels of training. Only 36 (31.9%) of the participating trainees expressed positive attitudes towards biostatistics. On the other hand, 30 (26.5%) participating trainees were found to have good biostatistics knowledge, compared to 83 (73.5%) trainees whose knowledge level was found to be poor. Upon adjusting for all background factors simultaneously, only younger age, level R4 training, publishing one or three papers were found to be associated with poorer attitudes towards biostatistics. Older age was associated with worsening of attitudes (adjusted odds = 0.9900, P = 0.00924), and so also was being a senior R4 trainee (adjusted odds = 0.9045, P = 0.01301). Publishing one paper (compared to publishing over three papers) was associated with poorer attitudes towards biostatistics (adjusted odds = 0.8857, P = 0.03525). Also, having published three papers (compared to publishing over three papers) was still associated with worse attitudes towards biostatistics (adjusted odds = 0.8528, P = 0.01318).

Conclusion:

The main finding of our current study is the poor level of knowledge and overtly negative attitudes held by family medicine trainees in Taif towards biostatics. Knowledge was particularly poor about advanced statistical concepts such as survival analysis and linear regression modelling. However, poor levels of knowledge about biostatistics could be a function of poor research productivity among family medicine trainees. Age, seniority in training and involvement in research also impacted positively on attitudes towards biostatistics. Therefore, it is recommended that the training curriculum for family medicine trainees should first cover essential biostatistics in a creative and accessible way and secondly encourage engagement research and publication from an early stage of training.

Introduction

Basic knowledge of statistical methods is an essential requirement for practice of safe evidence-based medicine.[1] Such statistics-related knowledge remains suboptimum, particularly among postgraduate trainees, making it a real challenge for postgraduate training schemes to manage.[2,3] Hence, a range of effective teaching methods were developed to improve knowledge of biostatistics and research methodology.[4] Clinical researchers often utilise statistical methods as they critique scientific papers and clinical guidelines or in peer review of papers submitted for publication.[5]

Despite its established importance, little is known about the knowledge of and attitudes towards statistics among trainees in family medicine, particularly in Saudi Arabia. Attitudes towards statistics were found to be not the best among medical trainees.[6,7] Biostatistics came with feelings of stress, frustration and insecurity among graduate medical students.[8] Statistics was considered by many a difficult subject.[9] Anxiety and scepticism were the dominating attitudes before engaging in statistical education among nurse students;[7] even worse is the fact that such negative and strong feelings did not lessen upon course completion.[9,10] Although such negative attitudes did not differ in accordance with sex or clinical rank, some factors impacted on them, such as specialism and mathematical background and statistical familiarity.[11] Medical teachers re-engineered their statistical curriculum delivery to help overcome negative attitudes by involving real-life practical sessions that could enhance students’ understanding of the value of biostatics. The preliminary results are promising.[12] Furthermore, clinicians with regular utilisation of research articles and advanced statistical concepts are found to be of higher knowledge, and therefore hold positive attitudes, of biostatistics.[13] Although Saudi-based physicians report positive attitudes towards biostatistics, they do realise the substantial gap in their knowledge and perception of statistical methods.[14]

Biostatistics is clearly an important subject, and its basic concepts need to be learned by practicing physicians and trainees. However, little is known about the knowledge[15] and attitudes[16] towards biostatistics among resident physicians in Saudi Arabia. Statistical knowledge among postgraduate medical residents in Jeddah was found to be quite poor.[17] This clearly corroborates results from around the globe.[18]

This current study aimed to estimate the level of knowledge and attitudes towards biostatistics among resident physicians in Saudi Arabia.

Materials and Methods

Study design: This study was a cross-sectional, questionnaire-based, descriptive study.

Study area: The study was carried out in Taif city.

Study population: The target population included all trainees in family medicine across their training level in Taif and Makkah regions.

Inclusion criteria: Residents currently in the training programme in family medicine in Taif, in the joint programme of the Armed Forces Hospitals and the Ministry of Health.

Exclusion criterion: Refusal to participate in the study.

Sample size: We aimed for complete enumeration of all family medicine residents in Taif.

Sampling scheme: We included all family residents in Taif, including those in the joint programme and those training within the Ministry of Health. The list of all family medicine trainees was provided by the Family Medicine Training Board. Our research tool covered aspects of attitudes towards statistics that included emotions, thoughts, preparedness of learning and perception of statistics value, in addition to clinical and demographic data. Data were entered into a spread Microsoft Excel sheet and stored on the hospital computer. Strict confidentiality was adhered to throughout the conduct of the survey.

Data collection tool: We utilised the underlying factors uncovered by previous studies conducted by Stanisavljevic et al.[10] and Alzahrani et al.[17] The questionnaire items were chosen to cover the factors of feelings about biostatistics, cognitive competence for biostatistics, value of biostatistics, difficulty of biostatistics and motivation to study biostatistics. All the items were re-examined in detail for construct validity and face validity by two senior family medicine consultants.

Regrading attitudes towards biostatistics, there were 26 items. Each comprised a statement about statistics; the trainee was required to rate his/her agreement with it between ‘strongly disagree’ and ‘strongly agree.’ An example of an item is,’ I like statistics’. Each item was given a score between 0 and 4, with higher score indicating favourable attitudes. Some items were reverse scored according to favourability of attitudes, for example, ‘statistics is worthless’. The total attitude score was calculated for the 26 questions, with a potential maximum score of 104. Any trainee who scored under 52 was deemed harbouring negative attitudes towards biostatistics.

In terms of knowledge about biostatistics, the participants were given a set of 15 statistical facts that the trainees had to rate as ‘correct’, ‘incorrect’ or ‘they do not know’. For each accurate response, a score of 1 point was awarded. If the trainee answered with ‘I don’t know’, then no points were given. The total knowledge score was calculated (potential maximum of 15 points). A score of 7 points or less meant the trainee had ‘poor knowledge’ about biostatistics.

Data collection method: Online questionnaire was sent to all residents in family medicine via email and social media accounts.

Data analysis: Demographic variables and background clinical characteristics of patients were represented with numbers and percentages and displayed as bar plots for categorical variables and as median, mean and interquartile range (IQR) and standard deviations (SDs) for continuous variables.

Effect of sociodemographic factors on attitudes towards biostatistics was assessed using multiple linear Poisson regression modelling.

All the analyses were conducted using R statistical software 3.6.0.[19]

The study was approved by the local research and ethics committee based in Al-Hada Armed Forces Hospital.

Results

The total number of trainees included in the study was 113 family medicine residents practicing in Taif, Saudi Arabia.

Only 36 (31.9%) of the participating trainees expressed positive attitudes towards biostatistics. The mean level of attitudes’ score towards biostatistics was 46.3 points (SD = 12.0 points), ranging between 9 points (minimum) and 84 points (maximum). The median attitude score was 48 points.

On the other hand, there were 30 (26.5%) participating trainees who were found to have good biostatistics knowledge, compared to 83 (73.5%) trainees whose knowledge level was found to be poor. The mean knowledge score was 5.1 points (SD = 4.2 points), ranging between 0 points (minimum) and 15 points (maximum). The median attitude score was 5 points.

Figure 1 shows that there were 30 (26.5%) participating trainees with good biostatistics knowledge, compared to 83 (73.5%) trainees whose knowledge level was found to be poor. As indicated in Figure 1b, only 36 (31.9%) of the participating trainees expressed positive attitudes towards biostatistics. Also, Table 1 shows the baseline demographic data and their unadjusted effect on knowledge and attitudes towards biostatistics.

Figure 1.

Levels of attitudes and knowledge about biostatistics among the participating family medicine trainees

Table 1.

Baseline demographic results of the participating family medicine trainees and the unadjusted effect on attitudes towards statistics

Factor	Count (n)/mean	%/SD	Mean/odds attitude score	Statistical test	P
Gender
Male	62	54.9%	47.5 points	t=−1.1998	0.2328
Female	51	45.1%	44.8 points
Age	Mean=28.1 years	SD=3.6 years	OR=0.9876	z=−3.843	0.0005
Nationality
Saudi	111	98.2%	46.3 points	t=0.1878	0.8811
Non-Saudi	2	1.8%	44.5 points
Training level
R1	26	23%	47.3 points	Reference	Reference
R2	26	23%	49.4 points	z=1.097	0.273
R3	27	23.9%	47.5 points	z=0.132	0.895
R4	34	30.1%	42.1 points	z=−2.958	0.003
Scientific publications
One	55	48.7%	45.9 points	Reference	Reference
Two	32	28.3%	48.5 points	1.759	0.0786
Three	19	16.8%	42.3 points	−2.013	0.0441
Above three	7	6.2%	50.0 points	1.517	0.1292

SD=standard deviation, r=correlation coefficient

Table 2 shows that, at the unadjusted level, only the number of publications influenced the knowledge score significantly. Trainees who published over three papers were far more knowledgeable than the rest (scored a mean of 8.4 points; P = 0.0001). Other attributes such as age, training level, gender or nationality did not have a significant impact on biostatistics knowledge.

Table 2.

Baseline demographic results of the participating family medicine trainees and the unadjusted effect on knowledge about statistics

Factor	Count (n)/mean	%/SD	Mean/OR knowledge score	Statistical test	P
Gender
Male	62	54.9%	5.2 points	t=−0.127	0.8992
Female	51	45.1%	5.1 points
Age	Mean=28.1 years	SD=3.6 years	OR=0.9818	z=−1.770	0.0767
Nationality
Saudi	111	98.2%	5.2 points	t=−1.9984	0.2417
Non-Saudi	2	1.8%	3.0 points
Training level
R1	26	23%	4.5 points	Reference	Reference
R2	26	23%	4.7 points	z=−1.107	0.268
R3	27	23.9%	5.5 points	z=0.091	0.927
R4	34	30.1%	4.9 points	z=−0.917	0.359
Scientific publications
One	55	48.7%	4.8 points	Reference	Reference
Two	32	28.3%	4.7 points	−0.269	0.7878
Three	19	16.8%	5.5 points	1.102	0.2703
Above three	7	6.2%	8.4 points	3.885	0.0001

OR=odds ratio, SD=standard deviation

Table 3 gives a detailed account of the results for each of the 26 attitude statements towards biostatistics among the participating family medicine trainees. Table 4 displays the individual responses of the participants to knowledge questions in the study.

Table 3.

Agreement with each of the 26 attitude statements towards biostatistics

	Attitude	Construct	Strongly agree		Agree		Neutral		Disagree		Strongly disagree
			n	%	n	%	n	%	n	%	n	%
1	I like statistics	Feelings	10	8.8	19	16.8	43	38.1	27	23.9	14	12.4
2	I enjoy taking statistics courses	Feeling	9	8.0	17	15.0	40	35.4	35	31.0	12	10.6
3	Statistics should be a required part of my professional training	Attitude	13	11.5	38	33.6	33	29.2	18	15.9	11	9.7
4	Statistical skills will make me more employable	Attitude	13	11.5	39	34.5	42	37.2	11	9.7	8	7.1
5	I use statistics in my everyday life	Attitude	9	8.0	14	12.4	35	31.0	33	29.2	22	19.5
6	Statistics formulas are easy to understand	Difficulty	5	4.4	20	17.7	39	34.5	33	29.2	16	14.2
7	Statistics is a subject quickly learned by most people	Difficulty	6	5.3	14	12.4	37	32.7	37	32.7	19	16.8
8	I am interested in teaching statistics	Interest	8	7.1	13	11.5	33	29.2	28	24.8	31	27.4
9	I am interested in using statistics	Interest	8	7.1	26	23.0	34	30.1	27	23.9	18	15.9
10	I am interested in learning statistics	Interest	11	9.7	35	31.0	32	28.3	19	16.8	16	14.2
11	I plan to work hard in my statistics course	Effort	12	10.6	28	24.8	56	49.6	8	7.1	9	8.0
12	I feel insecure when I have to do statistics problems	Feeling	16	14.2	42	37.2	36	31.9	11	9.7	8	7.1
13	I get frustrated going over statistics tests in class	Feeling	16	14.2	38	33.6	41	36.3	12	10.6	6	5.3
14	I feel under stress during statistics class	Feeling	16	14.2	42	37.2	34	30.1	17	15.0	4	3.5
15	I have trouble understanding statistics because of how I think	Cognitive	12	10.6	40	35.4	32	28.3	21	18.6	8	7.1
16	I make a lot of math errors in statistics	Cognitive	12	10.6	25	22.1	57	50.4	12	10.6	7	6.2
17	I find it difficult to understand statistical concepts	Cognitive	13	11.5	42	37.2	33	29.2	15	13.3	10	8.8
18	Statistics is worthless	Attitude	9	8.0	14	12.4	37	32.7	28	24.8	25	22.1
19	Statistics is not useful to the typical professional	Attitude	6	5.3	21	18.6	46	40.7	28	24.8	12	10.6
20	Statistics are rarely presented in everyday life	Value	11	9.7	28	24.8	38	33.6	30	26.5	6	5.3
21	Statistics is irrelevant in my life	Value	10	8.8	23	20.4	40	35.4	30	26.5	10	8.8
22	Statistics is a complicated subject	Difficulty	16	14.2	48	42.5	31	27.4	17	15.0	1	0.9
23	Learning statistics requires a great deal of discipline	Difficulty	18	15.9	51	45.1	35	31.0	6	5.3	3	2.7
24	Statistics involves massive computations	Difficulty	15	13.3	40	35.4	49	43.4	6	5.3	3	2.7
25	Statistics is highly technical	Difficulty	18	15.9	42	37.2	44	38.9	7	6.2	2	1.8
26	Most people have to learn a new way of thinking to do statistics	Difficulty	18	15.9	44	38.9	42	37.2	7	6.2	2	1.8

Table 4.

Knowledge about biostatistics items and the responses of family medicine trainees

	Statistical statement	Yes		No		Don’t know
		n	%	n	%	n	%
C1	Age in years is considered a discrete variable	43	38.1	25	22.1	45	39.8
C2	Weight in kilogram is considered a continuous variable	65	57.5	17	15.0	31	27.4
C3	Sex is considered a binary variable	56	49.6	12	10.6	45	39.8
C4	To study the effectiveness of paroxetine for treatment of panic disorder, a randomised clinical trial can be conducted comparing the panic symptoms in 40 patients who took paroxetine with 40 patients who took placebo	54	47.8	12	10.6	47	41.6
C5	For comparing the mean age of onset between patients with bipolar disorder and healthy people, a t-test can be used	39	34.5	13	11.5	61	54.0
C6	For comparing the mean depressive score between six different ethnic groups of participants, a one-way ANOVA test is appropriate	31	27.4	17	15.0	65	57.5
C7	For comparing the percentage of women who became pregnant in three different IVF methods, a Chi-squared test is appropriate	25	22.1	21	18.6	67	59.3
C8	A P value is the chance of wrongly finding a difference between the experimental and the control groups	30	26.5	21	18.6	62	54.9
C9	For evaluating the effect of methotrexate on survival of lung cancer patients, a Kaplan-Meier analysis is appropriate	18	15.9	12	10.6	83	73.5
C10	When the 95% confidence interval for the odds ratio of the effect of age on probability of dying from COVID-19 is between 1.9 and 2.3, this means the result is statistically significant	27	23.9	12	10.6	74	65.5
C11	If a researcher wants to increase the power of their study, they can increase the sample size	50	44.2	17	15.0	46	40.7
C12	The sensitivity of a diagnostic test is the proportion detected by the test as positive among all diseased individuals	52	46.0	10	8.8	51	45.1
C13	Unadjusted estimates for the effects of demographic variables on the risk of contracting COVID-19 can be studied within simple regression analysis	25	22.1	13	11.5	75	66.4
C14	Adjusted estimates for the effects of demographic variables on the risk of contracting COVID-19 can be studied within multiple regression analysis	20	17.7	12	10.6	81	71.7
C15	95% of all observations fall within 2 standard deviations around each side of the mean of normally distributed data	43	38.1	8	7.1	62	54.9

ANOVA=analysis of variance, COVID-19=coronavirus disease 2019, IVF=in vitro fertilisation

The percentage of trainees who correctly identified the scope for Kaplan–Meier analysis was only 15.9%. Only 17.7% of trainees recognised the need for multiple regression for adjusted analysis, and only 22.1% recognised the requirement for simple regression in unadjusted analysis. Only 26.5% were able to correctly interpret the P value concept correctly.

Clearly, upon adjusting for all background factors simultaneously, only younger age, level R4 training and publishing one or three papers were associated with poorer attitudes towards biostatistics. Older age was associated with worsening of attitudes (adjusted odds = 0.9900, P = 0.00924), and so also was being a senior R4 trainee (adjusted odds = 0.9045, P = 0.01301). Publishing one paper (compared to publishing over three papers) was associated with poorer attitudes towards biostatistics (adjusted odds = 0.8857, P = 0.03525). Also, having published three papers (compared to publishing over three papers) was still associated with worse attitudes towards biostatistics (adjusted odds = 0.8528, P = 0.01318). A visual display of the adjusted estimates is shown in Figure 2.

Figure 2.

The adjusted effects, as estimated through Poisson regression modelling, for background factors on attitudes towards biostatistics among a sample of family medicine trainees in Taif, Saudi Arabia

Figure 2 and Table 5 show that older age was associated with worsening of attitudes (adjusted odds = 0.9900, P = 0.00924). This is clear from the steeply descending age line in Figure 2a. So also was being a senior R4 trainee (adjusted odds = 0.9045, P = 0.01301). This is also apparent in Figure 2b as R4 scored the lowest in terms of attitude score. Publishing one paper (compared to publishing over three papers) was associated with poorer attitudes towards biostatistics (adjusted odds = 0.8857, P = 0.03525). Also, having published three papers (compared to publishing over three papers) was still associated with worse attitudes towards biostatistics (adjusted odds = 0.8528, P = 0.01318). This is obvious from Figure 2e for the lower attitude score for both publishing ‘one’ and ‘three’ papers.

Table 5.

Adjusted effect for the background demographic factors on participants’ attitudes towards statistics measured by Poisson multiple regression modelling

Demographic/clinical factor	Adjusted odds	95% Confidence Interval for adjusted odds	P
Age	0.9900	0.9825-0.9975	0.00924**
Training level: R2	1.0346	0.9539-1.1221	0.41120
Training level: R3	1.0071	0.9278-1.0931	0.86649
Training level: R4	0.9045	0.8356-0.9791	0.01301*
Sex: male	1.0311	0.9728-1.0930	0.30193
Nationality: Saudi	0.9433	0.7594-1.1717	0.59786
Publications: one	0.8857	0.7911-0.9917	0.03525*
Publications: two	0.9570	0.8506-1.0768	0.46518
Publications: three	0.8528	0.7519-0.9672	0.01318*

* = p < 0.05; ** = p < 0.01; *** = p < 0.001

However, in terms of adjusted effect for background factors on knowledge about biostatistics, none of the factors was of significant effect, except for publishing over three papers. Having published over three papers was associated with far better knowledge odds (compared to publishing one paper only) (adjusted odds = 1.7741, P = 0.000086) [see Table 6 and Figure 3].

Table 6.

Adjusted effect for the background demographic factors on participants’ knowledge towards statistics measured by Poisson multiple regression modelling

Demographic/clinical factor	Adjusted odds	95% Confidence Interval for adjusted odds	P
Age	0.9798	0.9586-1.0014	0.066974
Training level: R2	0.9148	0.7122-1.1749	0.485402
Training level: R3	1.0755	0.8453-1.3684	0.553410
Training level: R4	0.9232	0.7323-1.1638	0.498955
Sex: male	0.9944	0.8359-1.1830	0.949524
Nationality: Saudi	1.5084	0.6631-3.4311	0.326943
Publications: two	0.9728	0.7895-1.1986	0.795535
Publications: three	1.1656	0.9224-1.4729	0.199431
Publications: more than three	1.7741	1.3327-2.3618	0.000086***

* = p < 0.05; ** = p < 0.01; *** = p < 0.001

Figure 3.

The adjusted effects, as estimated through Poisson regression modelling, for background factors on knowledge about biostatistics among a sample of family medicine trainees in Taif, Saudi Arabia

Figure 3 shows that having published over three papers was associated with far better knowledge odds (compared to publishing one paper only) (adjusted odds = 1.7741, P = 0.000086). This can be observed in Figure 3e as publishing ‘over three’ papers shoots up in terms of knowledge score. Notably, although Figure 3a shows a steep decline for the age line, there is a lot of variability as indicated by the surrounding blue lines (compare with Figure 2a, where the variability was much lower).

Discussion of Key Findings

The current survey included 113 residents of family medicine at different stages of their training. They all train in Taif in both Armed Forces and the Ministry of Health facilities.

The main finding of our current study is that the estimate for positive attitudes towards biostatics was quite low, at 31.9%, among the surveyed family medicine trainees. This negative attitude towards biostatistics seems to be a ubiquitous finding in the medical profession. Even graduate-entry medical students were found to regard statistics quite unpopular, with only 24% of them prepared to voluntarily take a biostatistics course.[8] There were only 44.7% of dental students with positive attitudes towards statistics, with a notable association between academic achievement and positive attitudes towards statistics.[20] Even trainees in plastic surgery who hold positive attitudes towards biostatistics were less confident when it came to knowledge about biostatistics.[21] Our study confirms that negative attitudes held during medical undergraduate years continue well into senior postgraduate medical training.

A recent study among nursing students indicated that attitudes towards statistics were not affected by background demographic factors.[22] However, upon adjusting for all background factors simultaneously among the participants in our study, younger physicians were found to hold more positive attitudes towards biostatistics. One survey among medical postgraduate students indicated that younger students (under 26 years of age) were more positive towards statistics than older students.[9] Senior R4 trainees, conversely, had better attitudes towards biostatistics than more junior trainees in our survey. This agrees with the findings of Zhang et al.,[9] as years in medical training correlated positively with attitudes towards biostatistics. Indeed, as students progress in medical training, the value of statistical methods becomes more sensible. Persistence of statistical training among medical trainees was shown to improve and maintain attitudes towards biostatistics.[23]

One interesting finding is that publishing one or three papers was associated with poorer attitudes towards biostatistics. This is counterintuitive, as research has shown that more involvement in academic work means more positive attitudes towards statistics.[9] It can be conceived that publishing a single paper would indicate limited exposure to statistical methods than publishing over three papers. However, it is difficult to explain why publishing three papers was associated with such negative attitudes towards statistics in our sample.

Knowledge of biostatics estimate was even worse. Only 26.5% of family medicine residents scored over half of the given questions rightly. This is obviously a very concerning finding. Poor knowledge of biostatistics would inevitably cause substantial difficulties in critiquing research findings and lead to inaccurate conclusions.[24] Our figure is close to the 27% estimate for confidence in biostatistics and research methods found at baseline among pharmacy residents surveyed by Barreto et al.[25] Similar poor knowledge about statistical concepts was uncovered among orthopaedics trainees, for instance, over 30% were unable to understand the implications of P value.[26] Even papers that reported better results in terms of statistical knowledge expressed grave concerns regarding confidence in use and application of statistical concept among trainees.[27] Luckily, persistent educational interventions were shown to improve both knowledge and confidence in using biostatics.[28] Our estimate for statistical knowledge clearly falls well behind the 73.6% figure found in an earlier Saudi-based investigation of clinicians’ knowledge about biostatistics.[14] However, our study did not rely on self-rating of knowledge about statistics. We asked direct questions about variety of topics in statistics that family medicine clinicians need to be familiar with.

A recent Jeddah-based survey attempted to assess knowledge about research and its methodology among a sample of primary healthcare clinicians.[29] Knowledge scores ranged between 19% and 37% among the participating clinicians. This agrees with our findings. We do believe that poor knowledge about biostatistics, as reflected in our current findings, would inevitably lead to poor knowledge about research methodology.

Looking specifically into which knowledge areas were performed poorly among our participating family medicine trainees, we found them to struggle badly with the concepts of regression modelling and survival analysis. These were the exact statistical concepts rated difficult by a sample of plastic surgery trainees and they performed them with least confidence.[21] Survival analysis was specifically challenging for paediatric trainees in terms of knowledge of its underpinnings and uses.[30] Our trainees struggled substantially with interpreting the concept of P value correctly. This is not unique to Saudi-based family medicine trainees though. Also, 63% of Argentinian clinicians had poor knowledge of what P values exactly were.[31]

Our results show that, at the adjusted level, only the number of publications influenced the knowledge score significantly. Trainees who published over three papers were far more knowledgeable than the rest. This again highlights how important it is to indulge into research and real-life use of biostatistics in improving knowledge and attitudes towards biostatics.[32] However, it can also be a two-way association. Better knowledge of biostatistics was shown to improve academic performance and research productivity.[33] Educational workshops were shown to improve knowledge of biostatistics among postgraduate trainees;[34] however, we confirm that better results would be attained by those who engaged actively in medical research. Recent interventions showed significant improvement in knowledge and confidence about biostatistics among trainees who were mentored into longitudinal research in addition to regular research teaching and workshop delivery.[25] Notably, other attributes such as age, training level, gender and nationality did not have a significant impact on biostatistics knowledge.

However, in terms of adjusted effect for background factors on knowledge about biostatistics, having published over three papers was associated with far better knowledge odds (compared to publishing one paper only) (adjusted odds = 1.7741, P = 0.000086). This result is a particularly worrying finding. Only a handful of clinicians engage in production of high-quality research. A recent survey found that only 20% of Saudi clinicians had published any academic paper in scientific journals.[29] Particularly, in family medicine, research productivity was found to be very poor over the last two decades.[35] There was very little year-by-year increase in research related to family medicine in Saudi Arabia.[35] Even among high-ranking universities in Saudi Arabia, under 4% of health-related papers were published in high-impact scientific journals.[36] We call for earlier involvement of family medicine trainees in research and publication. Recent studies confirmed that adopting ‘block design’ in teaching biostatistics has shown promising results in terms of improvement in learning and attitudes.[37] Postgraduate curriculum should provide hands-on training in research and biostatistics using real-life data collected by junior trainees during their clinical placements.[32] We may also conclude that poor levels of knowledge about biostatistics could be a function of poor research productivity among family medicine trainees.

The current survey possesses several strengths. We interviewed all the family medicine trainees in Taif. We assessed knowledge about biostatics by asking direct questions about statistical concepts and applications and offered the trainees to respond. We, therefore, avoided the trap of social desirability bias caused by self-rating of knowledge. However, several limitations should be acknowledged. The sample was in totality from Taif-based family medicine trainees. Therefore, it is difficult to generalise the results to other areas, regions or specialties.

Further research should focus on how to improve biostatics knowledge among family medicine trainees in terms of creative teaching methods in class and in field. Qualitative research would suit exploration of family medicine trainees’ barriers against learning and using biostatistics. Indeed, a special focus for research should be dedicated to the longitudinal follow-up of family medicine trainees throughout the 4-year training programme to explore how attitudes and knowledge of biostatistics vary.

Conclusion

Attitudes towards biostatistics were negative among family medicine trainees and knowledge was poor, particularly concerning advanced statistical concepts such as survival analysis and linear regression modelling. The main finding of our current study is the poor levels of knowledge about biostatistics that could be a function of poor research productivity among family medicine trainees. Age, seniority in training and involvement in research also impacted positively on attitudes towards biostatistics. Future research should evaluate educational and training interventions that could improve attitudes and knowledge of biostatics among family medicine trainees.

Recommendations

1. Training curriculum for family medicine trainees should first cover essential biostatistics in a creative and accessible way and secondly encourage engagement research and publication from an early stage of training.

2. We call for earlier involvement of family medicine trainees in research and publication. Postgraduate curriculum should provide hands-on training in research and biostatistics using real-life data collected by junior trainees during their clinical placements.

3. The best way to improve attitudes and knowledge of biostatistics is to get involved in real-life research conception, design, conduct, write-up and publication. Family medicine trainees should be encouraged to actively participate in research and should be appropriately rewarded with acceleration in training if they publish their research in high-quality journals.

Ethical approval

The approval to conduct the study was taken from the regional research and ethics team in Al-Hada Armed Forces Hospital, Taif.

Financial support and sponsorship

Nil.

Conflicts of interest

There are no conflicts of interest.