Surgical Endoscopy

, Volume 28, Issue 6, pp 1830–1837

Distribution of innate ability for surgery amongst medical students assessed by an advanced virtual reality surgical simulator

  • Andrea Moglia
  • Vincenzo Ferrari
  • Luca Morelli
  • Franca Melfi
  • Mauro Ferrari
  • Franco Mosca
  • Alfred Cuschieri
Article

DOI: 10.1007/s00464-013-3393-6

Cite this article as:
Moglia, A., Ferrari, V., Morelli, L. et al. Surg Endosc (2014) 28: 1830. doi:10.1007/s00464-013-3393-6

Abstract

Background

Surgery is a craft profession requiring individuals with specific, well-documented innate aptitude for manipulative skills. Yet in most countries, the current selection process of surgical trainees does not include aptitude testing for the psychomotor and manipulative skills of candidates.

Methods

A total of 125 participants (121 medical students and four expert surgeons) performed all 26 exercises of the da Vinci Skills Simulator, with six exercises being identified as metrics of aptitude for manipulative and psychomotor skills. The expert surgeons were enrolled as the control group to validate the performance of the most talented students.

Results

Eight students (6.6 %) significantly outperformed the remaining 113, obtaining a median value of the sum of weighted overall score on the six selected exercises of 52.7 % versus 21.0 % (p < 0.001). In contrast, 14 students (11.6 %) performed significantly worse and well below the performance of the other 107, with a median value of overall score of 8.7 % versus 24.1 (p < 0.001). There was no statistically significant difference between expert surgeons (control group) and the eight talented students (62.1 % vs. 52.7 %, respectively; p = 0.368). No significant correlation between exposure to video games and overall score (ρ = 0.330) was observed.

Conclusions

In terms of innate aptitude for manipulative and psychomotor abilities, the present investigation has documented two subpopulations that fall outside the norm for the group of medical students recruited for the study: (i) a small group (6.6 %) with a high level and (ii) a larger cohort (11.6 %) with low level (significantly below the norm) innate aptitude for surgery. Exposure to video game experience did not appear to influence performances on the da Vinci Skills Simulator.

Keywords

Innate aptitude Ability Psychomotor aptitude tests Selection of surgical trainees Surgical skills and competence da Vinci Skills Simulator 

Training residents in surgery is time-consuming and costly [1]. Hence, it is pivotal to recruit those candidates who are most likely to become technically gifted surgeons, providing good surgical care with optimal clinical outcome. The process used for the selection of surgical trainees traditionally has been largely based on cognitive attributes (academic achievements), and subjective assessment of personality, attitude, and motivation based on non-structured interview of shortlisted candidates [2, 3]. More recently, in the Republic of Ireland an objective assessment for the selection of surgical trainees was developed [4, 5, 6], although the long-term impact on the quality of surgical care resulting from this improved selection of surgical trainees is not yet available.

In any event, the objective of the selection process is to identify medically qualified individuals who, with training, will acquire the necessary level of proficiency required for independent safe surgical practice [7]. As selection for surgical training is ultimately based on medical students who have graduated in medicine, we considered it logical and appropriate to recruit, by open invitation, a large cohort of undergraduate medical students to address the question “are the innate attributes for surgery normally distributed amongst the medical student population, from which surgical trainees are ultimately selected?”

In most countries, with the exception of the Republic of Ireland [4, 5, 6], the current selection process for admission to surgical residency training programs ignores one of the essential components of surgical practice, namely that surgery is one of the ‘craft professions’. This is in sharp contrast to aptitude testing for the selection process in established use by other professions, e.g. aviation, space travel, and nuclear power industries [8]. In a Delphi study involving internationally recognized master surgeons from Europe and US, who were or had been directors of surgical training programs, all identified innate dexterity as the most important factor, preceding cognitive factors and personality traits as determinants of the ultimate technical proficiency a trainee surgeon acquires with training [9]. Innate dexterity reflects innate ability of an individual and is defined as ‘the adaptive capacity, trait, or attribute that a person brings to a given task’ [10]. Contrary to skills acquisition, it is thought that abilities are present at birth, do not change during life, and are not influenced by training activities. Ability differs from skill, which is a mix of innate abilities honed with training and practice in a particular task [11].

Aptitude can be measured by various psychometric tests, but the first reported system for the objective assessment of psychomotor ability was the Advanced Dundee Endoscopic Psychomotor Tester (ADEPT). Studies with ADEPT confirmed its ability to discriminate the level of the innate abilities of surgical residents for endoscopic tasks, and this study identified one trainee in the 1st year with diminished manual dexterity who was subsequently unable to complete the surgical curriculum and was advised to change career in the 3rd year [12].

Minimal access and, more recently, robotic surgery, require the right mix of psychomotor, visuospatial, and depth perception abilities to an even greater extent than open surgery [13, 14, 15]. The lack of a method for objective evaluation of performance during training for minimal access surgery has been the driver for R&D in virtual reality (VR) simulators, enabling the acquisition of quality metrics (overall score) and efficiency metrics (measurable physical variables such as time, path length, and force) [16, 17, 18, 19]. In particular, advanced VR simulators, exemplified by the da Vinci Skills Simulator (Intuitive Surgical, Sunnyvale, CA, USA), enable the documentation of progress by surgical trainees based on clearly defined performance metrics which provide an objective assessment of psychomotor and manipulative skills [20]. The da Vinci Skills Simulator has been tested for face, content, construct, concurrent, and predictive validity [21, 22, 23].

Materials and methods

Participating medical students

A cohort of 121 medical students from two universities in Italy without prior experience with the da Vinci Skills Simulator and surgical experience were recruited by unpaid, open, total, non-randomized invitation for participation in this prospective study. The cohort represented 16.7 % of the total student population of the universities (n = 633). The undergraduate medical training course in Italy is through a nationally standardized 6-year course. The recruited participants consisted of 15 from the undergraduate years (UGY) 1–3, and 44, 30, and 32 from UGY 4–6, respectively. The group comprised 70 males (60 %) and 51 females (42 %), with a median age of 23 years (range 18–32). Four expert robotic surgeons who use the da Vinci robot routinely in their practice were invited to participate in the study by performing the same tests on the simulator as the medical students. The expert robotic surgeons had a median age of 48 years (range 37–51) and had performed a median of 250 interventions with the da Vinci robot (range 100–350).

Aptitude testing

This consisted of a single session during which each student performed all 26 exercises provided by the da Vinci Skills Simulator, and was preceded by a standardized familiarization protocol consisting of a visit to an operating room dedicated to robotic surgery with the da Vinci surgical robot, followed by practical instructions on the use of the da Vinci Skills Simulator. The exercises of the da Vinci Skills Simulator (software release C60_P6_L1_B14) are grouped as EndoWrist manipulation 1 (n = 3), camera and clutching (n = 6), EndoWrist manipulation 2 (n = 5), energy and dissection (n = 4), needle control (n = 2), and needle driving (n = 6). The simulator was linked to a laptop through a frame-grabber (VGA2USB by Epiphan systems, Palo Alto, CA, USA), enabling video recordings of the simulator sessions of all participants.

Since the study was designed to assess the distribution of innate manipulative and psychomotor abilities, the medical students performed the exercises without watching the instruction videos. Following completion, a questionnaire obtaining impressions of the simulator was completed by each participant. An experienced engineer (AM) supervised the students and provided clarification when needed during the test.

Data processing

The overall score of the exercises takes into account the weighted average of several parameters, including time to completion, economy of motion, instrument collisions, excessive instrument force, instruments out of view, and master workspace range. The metrics of all 26 exercises of all participants were acquired and stored.

Six exercises were selected by an expert surgeon (AC) in innate ability for surgery as indices of aptitudes for innate dexterity, and manipulative and psychomotor abilities: three from the Camera and Clutching group (camera targeting 1, ring walk 2, and ring walk 3), one from the EndoWrist Manipulation 2 series (ring and rail 2), and two from the Needle Control group (needle targeting, and thread the rings). These exercises were selected since they test hand–eye–foot coordination, bimanual dexterity, and accurate manipulation (Table 1).
Table 1

Ability tested by the selected exercises

Exercise

Ability tested

Camera targeting 1

Accurately positioning of the camera while working in a large workspace, and practice keeping instruments in view during large camera movements

Ring walk 2

Accuracy in positioning the camera, and effective switch between camera control and dexterous manipulation of the surgical tools

Ring walk 3

Same goals as ring walk 2, but additional practice retracting obstacles using a third instrument

Ring and rail 2

Coordination of two-handed control of an object’s position and orientation along a trajectory using the surgical instruments

Needle targeting

Accurate manipulation and insertion of needles

Thread the rings

Two-handed manipulation with hands-off between instruments, and accuracy when driving a needle

Data analysis

Overall score, time of completion, economy of motion, and excessive applied force were computed for the six selected exercises by applying normalized weighting factors. The sum of the weighted average of the overall scores and the other parameters on the six selected exercises provided the performance of each student.

We defined high performers as those students whose weighted overall score was at least double that of the remaining ones. Likewise, we identified as under performers those with a weighted overall score less than half of the other ones.

As the data were non-parametric with skewed distribution, analysis was performed by Wilcoxon signed-rank test (p < 0.001) using IBM SPSS Statistics 22 (Chicago, IL, USA), on the overall score (quality metrics) and the three main efficiency metrics: time of completion, economy of motion, and excessive force. Correlation test was assessed using the Spearman coefficient (ρ).

Results

The median time to complete a session of all 26 exercises was 144 min (range 99–247). The scores for the six selected exercises for all 121 students are shown in Table 2. The median of the overall score on these tasks was: 44 % (Q1 30, Q3 58) for camera targeting 1; 32 % (Q1 18, Q3 50) for ring walk 2; 2 % (Q1 0, Q3 13) for ring walk 3; 26 % (Q1 18, Q3 35) for ring and rail 2; 58 % (Q1 31, Q3 74) for needle targeting; and 42 % (Q1 27, Q3 55) for thread the rings. These results suggest that ring walk 3 is the most exacting exercise with the lowest overall score due to long execution time to complete the exercise (median of 374.2 s) and poor economy of motion (514.5 cm), measured as the total path generated by the virtual robotic arms. As in previous studies, execution time and economy of motion correlated positively and are used as indices of surgical technical ability and efficiency [13]. The particularly complex task scenario of ring walk 3 was accompanied by the highest incidence of instrument collisions (median of 22) and time in which the instruments were out of view (49 s). Ring and rail 2 is the exercise that requires the greatest precision and manipulative abilities. Not surprisingly, it had the highest mean execution time (median of 427.7 s) and economy of motion (633.7 cm). The scores of ring walk 3 and ring and rail 2 were also accompanied by excessive force by the users, indicated by the robotic arms becoming red, with a median of 87 and 81 s, respectively.
Table 2

Metrics data of the six selected exercises for all medical students (median, interquartile 1, and interquartile 3)

 

Camera targeting 1

Ring walk 2

Ring walk 3

Ring and rail 2

Needle targeting

Thread the rings

Overall score (%)

44.0, 30.0, 58.0

32.0, 18.0, 50.0

2.0, 0.0, 13.0

26.0, 18.0, 35.0

58.0, 31.0, 74.0

42.0, 27.0, 55.0

Time to complete exercise (s)

165.0, 131.2, 235.9

259.9, 213.3, 327.6

374.2, 303.2, 495.4

427.7, 357.2, 535.5

336.9, 264.4, 425.5

235.4, 203.7, 280.1

Economy of motion (cm)

196.2, 158.8, 246.0

308.8, 247.8, 434.5

514.5, 419.0, 780.9

633.7, 529.2, 804.8

410.7, 301.8, 587.2

328.9, 272.7, 415.5

Excessive instrument force (s)

0.0, 0.0, 3.0

14.0, 1.5, 38.1

87.0, 38.0, 191.0

81.0, 36.5, 157.0

2.5, 0.5, 12.5

9.0, 2.5, 22.5

The mean overall score of the students on the six selected exercises was used to apply a weighting factor to each exercise. This was 43.80 % for camera targeting 1, 34.67 % for ring walk 2, 8.47 % for ring walk 3, 28.32 % for ring and rail 2, 52.37 % for needle targeting, and 41.84 % for thread the rings. A normalized weighting factor, computed as illustrated in Fig. 1, was then assigned to each task as follows: 9.20 % for camera targeting 1, 11.63 % for ring walk 2, 47.60 % for ring walk 3, 14.23 % for ring and rail 2, 7.70 % for needle targeting, and 9.64 % for thread the rings.
Fig. 1

Algorithm used to compute the normalized weighting factors associated with the six selected exercises

The frequency histograms of the weighted overall score with fitted curves of the data distribution for the six selected exercises, illustrated in Fig. 2, show that most students obtained a weighted score ranging from 10 to 30 %, with few exceeding 50 % and some below 10 %. Figure 3 shows the different weighted overall scores for the three subpopulations (least talented, average, and most talented medical students). In particular, there is no overlap among the performances of the three groups.
Fig. 2

Frequency histogram of the weighted overall scores with fitted curves of the data distribution for the six selected exercises

Fig. 3

Weighted overall scores for the three subpopulations: least talented, average, and most talented medical students

Generally, males performed better than females, with a median of 26.7 % (Q1 17, 3, Q3 32.6) and 18.9 % (Q1 14.8, Q3 24.3), respectively, as illustrated in Fig. 4.
Fig. 4

Weighted overall scores for male and female participants

The questionnaire following completion of the exercise showed that irrespective of talent, the majority of students rated the simulator as very intuitive (8.4 ± 0.9 out of 10) and with good visual realism (7.8 ± 1.2 out of 10).

High performers (talented students)

By applying the normalized weighting factors (Fig. 1), we found eight students (6.6 %), consisting of seven males and one female, who significantly outperformed the other 113, obtaining an overall score more than double that of the others of 52.7 % (Q1 50.4, Q3 56.2) and 21.0 % (Q1 15.2, Q3 28.5), respectively (Table 3). With respect to parameters that are acknowledged as strong predictors of surgical ability, i.e. completion time and economy of motion, the results for the top eight students (based on the metrics of the da Vinci Skills Simulator) were 269.5 s (Q1 254.7, Q3 306.5) versus 349.2 s (Q1 304.1, Q3 433.0) [p < 0.001] for time to complete the exercise, and 293.1 cm (Q1 269.1, Q3 334.2) versus 470.2 cm (Q1 397.7, Q3 680.1) [p < 0.001] for economy of motion.
Table 3

Performance of the most talented students versus the rest for overall score, time to complete, economy of motion, and excessive applied force on the six selected exercises (median, interquartile 1, and interquartile 3)

 

Overall score (%)

Time (s)

Economy of motion (cm)

Excessive force (s)

Most talented students (n = 8)

52.7, 50.4, 56.2

269.5, 254.7, 306.5

293.1, 269.1, 334.2

5.4, 2.2, 24.9

Remaining students (n = 113)

21.0, 15.2, 28.5

349.2, 304.1, 433.0

470.2, 397.7, 680.1

70.9, 38.6, 146.0

p < 0.05

<0.001

0.004

<0.001

<0.001

Results obtained using normalized weighting factors

They also avoided using excessive applied force during the execution of the tasks compared with the rest: 5.4 s (Q1 2.2, Q3 24.9) versus 70.9 s (Q138.6, Q3 146.0) [p < 0.001]. Five were from UGY4, two from UGY5, and one from UGY6. The study identified a subgroup of students with outstanding manipulative skills comprised of 7 out of 70 male students (10 %) and 1 out of 51 female students (2 %).

The performance of the four expert robotic surgeons contrasted with the eight best students is shown in Table 4. Although the performance by the expert surgeons was better, the difference was not significant with respect to overall score [62.1 % (Q1 55.2, Q3 67.5) versus 52.7 % (Q1 50.4, Q3 56.2)], economy of motion [271.5 cm (Q1 251.7, Q3 292.2) versus 293.1 cm (Q1 269.1, Q3 334.2)], and excessive force [4.3 s (Q1 3.3, Q3 6.5) versus 5.4 s (Q1 2.2, Q3 24.9)]. However, as expected, the expert surgeons outperformed the eight students with regard to completion time: 121.1 s (Q1 117.4, Q3 124.1) versus 269.5 s (Q1 254.7, Q3 306.5) [p = 0.004].
Table 4

Performance of the expert surgeons on the da Vinci robot simulator versus that of talented students (median, interquartile 1, and interquartile 3)

 

Overall score (%)

Time (s)

Economy of motion (cm)

Excessive force (s)

Expert surgeons (n = 4)

62.1, 55.2, 67.5

121.1, 117.4, 124.1

271.5, 251.7, 292.2

4.3, 3.3, 6.5

Talented students (n = 8)

52.7, 50.4, 56.2

269.5, 254.7, 306.5

293.1, 269.1, 334.2

5.4, 2.2, 24.9

p < 0.001

0.368

0.004

0.368

0.933

Results obtained using normalized weighting factors

Under performers (below norm) students

In total, 14 students (11.6 %), consisting of seven males and seven females, performed significantly worse than the other 107, obtaining a median weighted overall score less than half of the remaining of 8.7 % (Q1 6.8, Q3 10.3) and 24.1 % (Q1 18.2, Q3 30.8), respectively (Table 5). The completion times of this group were significantly higher [516.5 s (Q1 455.4, Q3 614.5)] than the majority [334.0 s (Q1 287.9, Q3 399.0)], as were the economy of motion [810.5 cm (Q1 741.3, Q3 1139.9) versus 435.5 cm (Q1 366.5, Q3 551.0)], and the excessive force applied during execution [183.3 s (Q1 122.6, Q3 228.8) versus 57.3 s (Q1 32.3, Q3 94.7)] (p < 0.001).
Table 5

Performance of the least talented students versus the rest for overall score, time to complete, economy of motion, and excessive applied force on the six selected exercises (median, interquartile 1, and interquartile 3)

 

Overall score (%)

Time (s)

Economy of motion (cm)

Excessive force (s)

Least talented students (n = 14)

8.7, 6.8, 10.3

516.5, 455.4, 614.5

810.5, 741.3, 1139.9

183.3, 122.6, 228.8

Remaining students (n = 107)

24.1, 18.2, 30.8

334.0, 287.9, 399.0

435.5, 366.5, 551.0

57.3, 32.3, 94.7

p < 0.05

<0.001

<0.001

<0.001

<0.001

Results obtained using normalized weighting factors

The distribution within years of the curriculum of these least dexterous students consisted of one each from UGY1–3, six from UGY 4, two from UGY 5, and three from UGY 6. The results indicate that 10 % (7 out of 70) of male medical students and 13.7 % (7 out of 51) female medical students have a below-average aptitude for manipulative skills.

Influence of video game experience

Overall, 56 (46.3 %) students had no or little previous experience with video games, while 28 (23.1 %) had less than 10 years’ experience [median 5.5 years (Q1 3.5, Q3 7.0)], and 37 (30.6 %) played video games for a period ranging from 10 to 20 years [median 12 years (Q1 10, Q3 15)].

The non-gamers obtained a median weighted overall score of 19.8 % (Q1 14.5, Q3 24.6) versus 22.1 % (Q1 15.1, Q3 28.5) and 30.2 % (Q1 18.9, Q3 37.5) of those with less and more than 10 years’ experience (Fig. 5). Although these results may suggest that performances at the da Vinci Skills Simulator increases with video game experience, a Spearman test showed a low non-significant correlation (ρ = 0.33).
Fig. 5

Weighted overall scores with respect to the level of experience in video games: no experience, less than 10 years (low), and more than 10 years (high)

Discussion

Contrary to intelligence quotient, which measures a number of attributes such as verbal comprehension, working memory, perceptual organization and processing speed, aptitude refers to certain characteristics of mental ability which define the adeptness level of individuals for certain tasks. Thus, aptitude is variously defined as innate specific ability, which facilitates competent performance of a task, an innate capacity for a particular activity, or, in some instances, an innate component of a defined competency [2, 3, 4, 5, 6, 7, 10, 19]. Aptitude assessment is used for vocational/career guidance and planning, and for recruitment into various professions (including aviation, space travel and nuclear power industries) by a range of psychometric tests that measure different aptitudes, e.g. general learning ability, numerical ability, verbal ability, spatial perception, psychomotor dexterity, etc. [8, 24]. Aptitude has to be distinguished from attitude, which implies strong commitment and motivation of an individual to excel in a particular activity, and which can be measured by tests such as the employment inventory. Both are important in the achievement of competency in a given profession.

Aptitude tests attempt to determine the individual’s ability to acquire, through training, a certain specific set of skills (intellectual, motor, etc.) needed for various professions. These aptitude tests assume that individuals vary in their special innate abilities, comprising three populations: (i) the norm with average ability constituting the majority; (ii) the talented with respect to the activity (top percentile); and (iii) the unsuited (with respect to the activity) with aptitudes below the norm. These differences between individuals are useful in predicting future achievements, and thus highly relevant to the craft professions such as surgery.

Several studies on one of the early VR simulators for flexible endoscopy (GI Mentor by Simbionix; Cleveland, Ohio, USA) and laparoscopic surgery (MIST-VR surgical simulator by Mentice; Gothenburg, Sweden) indicated that visuospatial and psychomotor skills correlated to the number of sessions needed to train subjects [17, 25, 26, 27, 28, 29, 30].

In the present study, 125 participants performed 3,250 exercises at the da Vinci Skills Simulator. This is the largest single-center reported study on the da Vinci Skills Simulator at the time of this writing. This work has addressed the distribution of psychomotor skills amongst medical students using an advanced validated VR simulator, and identified two outlying groups (above and below the norm) in terms of manipulative and psychomotor skills relevant to surgery, with the former exhibiting performance on the da Vinci Skills Simulator that was comparable to that by a control group of expert surgeons [19]. The present study confirmed an unexpected finding, namely that the population of medical students who are least talented in terms of manipulative and psychomotor abilities required for surgery is almost double the population who are specially gifted for this craft profession (11.6 % vs. 6.6 %). Thus, for practical purposes if and when objective aptitude testing becomes universally included in the selection process of surgical trainees, the identification of those who are manifestly unsuited is more important than identification of talented individuals, especially as other factors, including motivation, communications, team spirit, personality type, etc..., influence the level of competency achieved by training.

Although previous reports documented a positive impact of video games in laparoscopy, both when using VR simulators and performing surgical tasks, a review concluded that they do not enhance robotic surgical performance [11, 31, 32, 33]. This study confirmed a low correlation between video game experience and a VR simulator for robotic surgery.

This study has one limitation that we acknowledge—the intake into the study was not randomized but was instead by total open unpaid invitation. We opted for this design as the study needed a large cohort of participants. However, we do not regard this as a major limitation as there was no attempt at selection of the medical students, although we admit to the possibility of self-selection. This is unlikely to be the case from the results obtained as the poor performers outnumber the top performers by a factor of almost 2.

Conclusions

The advent of surgical simulators offers an opportunity not just to change to a competence-based curriculum, but also to use VR systems for aptitude testing in the selection of surgical trainees to complement academic achievement, personality, and motivation. On the basis of several recent reports, the case for aptitude testing in the selection of surgical trainees is overwhelming and the technology is now available for this to be carried out objectively. The present study has shown that distribution of the innate manipulative and psychomotor skills for surgery has two extreme groups—a small group (7 %) with innate aptitude for these skills that is way above the norm, and a larger group (12 %) of poor performers with aptitude significantly below the norm. If we assume that these students’ aptitude levels remain substantially unchanged when they qualify and complete their internship, then without a rigorous selection process the odds for selecting surgical trainees with low levels of innate aptitude for surgery extrapolated from the present study would be substantial.

Acknowledgments

The authors would like to thank staff at the Multidisciplinary Robotic Center at the Cisanello University Hospital of Pisa for their assistance during the tests.

Disclosures

Drs. Andrea Moglia, Vincenzo Ferrari, and Luca Morelli, and Professors Mauro Ferrari, Franco Mosca, and Alfred Cuschieri have no conflict of interests. Dr. Franca Melfi is Official Proctor at the Multidisciplinary Center of Robotic Surgery at Cisanello Hospital in Pisa, Italy.

Funding

Study partly funded by Regione Toscana.

Copyright information

© Springer Science+Business Media New York 2014

Authors and Affiliations

  • Andrea Moglia
    • 1
  • Vincenzo Ferrari
    • 1
  • Luca Morelli
    • 1
    • 2
  • Franca Melfi
    • 2
  • Mauro Ferrari
    • 1
  • Franco Mosca
    • 3
  • Alfred Cuschieri
    • 4
    • 5
  1. 1.EndoCAS, Center for Computer Assisted SurgeryUniversity of PisaPisaItaly
  2. 2.Multidisciplinary Center of Robotic SurgeryUniversity Hospital of PisaPisaItaly
  3. 3.Cisanello University Hospital of PisaPisaItaly
  4. 4.Scuola Superiore Sant’Anna of PisaPisaItaly
  5. 5.Institute for Medical Science and TechnologyUniversity of DundeeDundeeUK

Personalised recommendations