Reference letters for subspecialty medicine residency positions: are they valuable for decision-making? Results from a Canadian study

The letter of recommendation is currently an integral part of applicant selection for residency programs. Internal medicine residents will spend much time and expense completing sub-specialty away electives to obtain a letter of recommendation. The purpose of this study was 1) to examine a large sample of reference letters in order to define essential components of a high-quality letter, and 2) to elucidate the relationship between quality of reference letter and the letter writer. We conducted a two-phase study. In phase one, a large sample of letters of recommendation was examined using an audit tool as a coding framework. A 5-point composite endpoint of high-quality letter components was subsequently developed. In phase two, program director letters were compared to non-program director home institution and non-home institution elective letters based on inclusion of components of the 5-point composite endpoint using Chi square testing. 715 letters were examined (398 non-program director home institution letters, 201 program director letters, and 116 non-home institution elective letters). High-quality letter components were: nature of relationship, duration of relationship, In Training Evaluation Report information, research involvement and comments on areas for improvement. Program director letters had a significantly higher proportion (10.4%) of all 5 high-quality components, compared to 0% in both non-program director home institution letters and elective letters (p < 0.001). A significantly higher proportion of program director letters had 4–5 high-quality components (62.5%) compared to 2% of non-program director home institution letters and 0% of elective letters (p < 0.0001). Letters of recommendation from elective rotations are of the poorest quality and such rotations should not be pursued for the sole purpose of obtaining a letter. The low quality of elective letters leads to the recommendation that writers should decline to write them, programs should not require them and trainees should not request them. Program directors write the highest quality letters and should be a resource for faculty development. Clinical supervisors can use the 5-point composite endpoint as a guide when writing letters for applicants.


Background
The Letter of Recommendation (LOR) is an integral part of applicant selection for internal medicine subspecialty training programs. LORs are considered one of the most important factors in ranking candidates to postgraduate subspecialty programs and may be helpful in predicting residency retention rates [1][2][3][4][5]. However, the writing of LORs has also been described as a process that very few understand [6] and the role of the LOR has been deemed worthy of further examination.
LORs have been found to lack a meaningful comparison of the applicant to peers and often fail to include concrete examples demonstrating applicant performance [7]. LORs will often include "lengthy reiterations of already available objective data" and render all applicants "excellent," making it extremely difficult to discriminate between them [8,9]. In response to these noted shortcomings, program directors and medical educators are often asked to comment on what could make LORs more helpful. Researchers have examined, in a limited fashion, the role of the letter writer [10], key elements and phrases in letters and correlations between LORs with objective data [11,12].
Our previous research which included a survey of Canadian internal medicine program directors has provided some insight into the challenges with letter writing and interpretation. Program directors (PDs) felt LORs lacked a common vocabulary and format and that referees often used variable rating scales or similar statements that did not differentiate between candidates. One example of this is the observation that a significant proportion of residents appeared to be in the top 5% of applicants [13]. The Canadian internal medicine PDs also suggested that there may be certain reference letter features that are more useful than others, and this led us to explore LORs further [13].
We partnered with the Canadian Resident Matching Service (CaRMS) to evaluate LORs for internal medicine residents applying to subspecialty training programs. The study objectives were to: 1a) elicit the key components of letters which affect quality through a large audit of LORs to subspecialty internal medicine programs in Canada between 2011 and 2014 and b) subsequently define the components of a high-quality letter; and 2) compare PD letters to non-PD home institution and non-home institution elective letters (elective letters) based on the inclusion of high-quality components.

Methods
Ethics approval was obtained from the Queen's University Health Sciences & Affiliated Teaching Hospitals Research Ethics Board and the Canadian Resident Matching Service (CaRMS) Board.
This study was designed in two phases.

Phase 1: creation of composite endpoint for a highquality LOR
The objective of phase one was to identify common components present in LORs and then by consensus, select the components that are necessary for a LOR to be deemed high-quality.  [14,15] using open coding of the reference letters with the audit tool as a coding framework to identify common components. She independently reviewed 10% of reference letters. The research team subsequently reviewed the same data set. Discrepancies among the team were resolved and the audit tool was refined for clarity. The resultant iteration of the audit tool provided a clear list of components for ongoing coding (Appendix). The research assistant was not associated with any resident or faculty investigators to address reflexivity and mitigate bias [16].
Defining a high-quality letter: construction of the composite endpoint A composite endpoint consisting of five items was created to define a high-quality LOR. The composite endpoint was created by selection of the five most valuable components of LORs from the list of identified common components, based on: 1) the opinions of Canadian internal medicine program directors obtained through a survey [13] (Fig. 1), 2) results of a detailed literature review, and 3) consensus from a panel of local medical educators, including subspecialty medicine program directors and the dean of postgraduate education.

Phase 2: reference letter comparison based on referee type
Letters from PDs (PD letters) were compared to (i) letters by non-PDs from the home school (non-PD home institution letters), and (ii) letters obtained on elective rotations (elective letters). Letters were compared with respect to the inclusion of individual components of the 5-point composite endpoint. The inclusion of In Training Evaluation Report (ITER) information was identified as one of the components of the composite endpoint. Upon reflection that elective letters were unlikely to include this data, a separate comparison was also performed, removing the ITER component and leaving a 4point composite endpoint. Analysis was also performed to assess if a writer had met the criteria for a good to high-quality letter by including a majority of the components.
Chi square testing was performed to compare frequencies using SPSS Statistics, version 24 (IBM, Armonk, New York). A sample size of 105 per group was necessary to demonstrate a 20% difference in proportion of high-quality letters using three-step pairwise comparison in an ANOVA.

Phase 1: creation of composite endpoint for a highquality LOR
The components necessary for a high-quality letter were determined to be: (1) nature of relationship between the referee and candidate, such as rotation supervisor, research supervisor, and/or program director (2) duration of this relationship-measured in weeks.
(3) inclusion of In Training Evaluation Report (ITER) information (4) research involvement comments, often outlining role in specific projects (5) area for improvement or critique comments.
Given the possibility that only PDs would have ready access to ITER information, we also performed the analysis excluding this component, using a 4-point composite endpoint.
Phase 2: reference letter comparison based on referee type 738 letters were reviewed, with 715 included in the final analysis. Twenty-three were excluded due to duplicates or redaction beyond ability to extract meaningful information. The distribution of reference letters was: 398 non-PD home institution letters (55.7%), 201 PD letters (28.1%), and 116 elective letters (16.2%). Letters were written for 206 applicants, with a range of 1-5 letters per applicant reviewed. Applicants were applying for 9 specific sub-specialties, with good representation of the breath of sub-specialty internal medicine. There were 113 letters for which the specialty being applied for was not evident. There were no letters of application for allergy, infectious disease, or endocrinology. Letter writers were affiliated with 33 different institutions within Canada and the United States. With respect to the primary outcome (inclusion of all 5 high-quality components of the 5-point composite endpoint), PD letters had a significantly higher proportion of letters that met this target (10.4%), compared to 0% in both non-PD home institution letters and elective letters (p < 0.001) ( Table 1). Using the "non ITER" 4point composite endpoint, PDs again had a significantly higher proportion with all four components (12.4%), compared to non-PD home institution letters (1.3%) and elective letters (0%) ( Table 2).
When evaluated for having a majority of the components of the 5-point composite endpoint, a significantly higher proportion of PD letters had 4-5 components (62.5%) compared to 2% of non-PD home institution letters, and 0% of elective letters (p < 0.0001) ( Table 1). Of PD letters, 94% contained 3 or more components, compared to 37.7% of non-PD home institution letters, and 11.2% of elective letters (p < 0.0001) ( Table 1). Likewise, for the 4-point composite endpoint, 73.1% of PD letters had 3-4 components compared to 36.9% of non-PD home institution letters and 10.3% of elective letters (p < 0.0001) ( Table 2). Further breakdown of letter components is seen in Table 3.

Discussion
Letters of recommendation continue to be a key component of the modern fellowship application [6]. However, this current study of over 700 LORs reveals a significant variability in letter quality. Variability in letter quality appeared dependent on the relationship between the writer and applicant. Not surprisingly, PDs' letter writing practice is consistent with their recommendations. This is demonstrated by the fact that PD letters had the highest proportion of high-quality letters (10.4% included all 5 components of the 5-point composite endpoint and 94% had 3 or more components), while letters from elective supervisors were of much lower quality, with 0% having all 5 components of the 5-point composite endpoint, and only 11.2% having 3 or more components.
Although there is existing literature examining LORs, the current study adds to this in several respects. Most studies to date have looked at trainees in the United States medical training system [10,11,17], while the current study examined letters from across Canada, with a different healthcare and education system, thus adding the experience of another country and enriching the data pool. This study also highlights the relationship between letter quality and letter writer, leading to opportunity for possible faculty development with a goal of improving letter quality. Lastly, it brings forth the issue of elective rotations, and their role as auditions for position, and whether letters of recommendation should be a part of key outcomes of this effort, particularly in light of the poor quality of the letters that come from the experience.

The letter writer
Program directors practice what they preach. They understand the value of specific components of LORs and include these components with the highest frequency. This indicates that PDs are an untapped resource for providing faculty development to all clinical supervisors who are potential letter writers. In stark contrast to PD letters, LORs from elective rotation supervisors were of the lowest quality. One possible reason for this finding is that a short amount of time spent with an unknown trainee does not provide sufficient data to allow the supervisor to be comfortable in rating the trainee in key domains. As many residents will seek out elective rotations at other sites with the intention of not only introducing themselves, but also to obtain a LOR from this site, the overall poor quality of these letters calls into question at least one element of the perceived value of these rotations.

Elective rotations
While letters from elective supervisors represent approximately 15% of overall letters in our sample, their utility appears very limited; a meagre 11.2% met the criteria for a good to high-quality letter by including 3 or more components of the 5-point composite endpoint. Questions raised by this include 1) why would trainees get these letters, and 2) did this study miss some important aspect of the letter that would make letters from electives important? While our data suggest elective letters are not of good quality, and thus seemingly not useful overall, they may be a necessity for a particular institution, regardless of how non-specific the information within them, as suggested by recent papers from the surgery field [18,19]. These could be considered letters from an 'audition elective'-a term used to describe a clinical elective taken by trainees to distinguish themselves from their peers in order to improve their chances of being selected by a residency program [20]. Internal medicine residents will spend a great amount of time and expense completing sub-specialty audition electives to improve their chances of success in the subspecialty match, possibly via obtaining a strong reference letter from their elective supervisors. As we have shown the LORs from elective rotations to be of low quality, this calls into question the value of these rotations, given their associated costs. With the current travel restrictions related to the COVID-19 pandemic, completion of these away rotations may be even more difficult. This is consistent with other studies questioning the value of the 'away rotation' and that the value of such rotations should not depend solely on the obtaining of a LOR [21]. The current study does have certain limitations. While exposing key differences in letter quality according to the letter writer, further exploration of this data is somewhat limited. To ensure anonymity, the letters underwent redaction prior to release by CaRMS, with demographic data including age, gender, and candidate home institution name being removed. Moreover, we were unable to correlate these findings with subspecialty match or residency performance results, thus limiting us to examining the process rather than outcomes that may be of interest to a PD. We put a lot of weight on the opinions of internal medicine PDs, largely because of their experience in both writing and reading letters of recommendation. However, there are other members of selection committees, whose views and opinions of a quality letter may differ. In consideration of this, our composite endpoint was developed with input from the internal medicine PD survey, but also with input from a panel of academic internal medicine subspecialists who sit on various residency training selection committees. Lastly, the current study only looked at letters for application to the subspecialty internal medicine match and results may not be transferable to programs accepting medical students into their first year of residency training (PGY-1).

Recommendations
(1) LORs from elective rotations: We recommend that letters from elective rotations should not be mandatory for consideration of acceptance into a program, as well as consideration of a faculty wide approach/position that writers will not provide letters based solely on elective rotations. This recommendation would presumably lead to trainees asking for letters from other sources, likely faculty that know them better and thus can provide a higher quality letter.