About Dataset

Introduction

Shoulder injuries are among the most common types of upper extremity injuries in both contact and noncontact sports. They are a significant source of morbidity for athletes, accounting for almost one third of all sports-related injuries (Enger). As a result of these factors, shoulder injuries and their post healing metrics are an important area for research in orthopaedics.

Shoulder injuries most commonly result from direct trauma or a fall onto the ipsilateral shoulder, making athletes especially prone to these types of injuries (Monica). Some of the most common injuries in this population include anterior and posterior glenohumeral instability, acromioclavicular pathology (including separation, osteolysis, and osteoarthritis) and rotator cuff tears (Gibbs). Acromioclavicular joint injuries are the most common among the athletic population with an overall incidence rate of 9.2 per 1000 person-years and an average time of 18.4 days lost per athlete (Pallis) followed by Glenohumeral instability at at 2.79 per 1000 person-years (Lanzi).

With American football being a high contact sport played at high speeds, the potential for shoulder injuries from minor sprains to career ending tears is significant. Nearly half of players at the NFL combined have reported a history of shoulder injury, with 34% requiring operative intervention (Kaplan). Quarterbacks are particularly affected by shoulder injuries due to their playing position being targeted by the opposing team on every play, and the associated throwing mechanics with their playing actions. Of all QB injuries reported, shoulder injuries are the 2nd most common at 15.2% (Kelly).

The purpose of this study was to determine (1) the general impact on performance metrics among NFL quarterbacks following shoulder injury and (2) the impact of surgical interventions to repair these injuries had on career outcomes using measures such as passer rating, yards ran, and successful passes. We hypothesized that quarterbacks in the national football league who injure their dominant shoulder will 1) have decreased performance metrics after surgery 2) those that get surgery will have better performance metrics compared to those that do not get surgery.

Methods

Overview

National Football League (NFL) Injured Reserve (IR) lists for the years 1980 to 2019 were pulled from Pro Sports Transactions and entries were queried to find quarterbacks who were placed on the IR with a shoulder injury.

50 quarterbacks were found to have long-term shoulder injuries, and a subset of these were selected who had first-time shoulder surgery on their dominant, throwing arm. Manual searches were performed to verify the nature of the injury and determine dates of surgery. Age-matched controls were selected with the following criteria:

same years of experience
same number of career seasons +/- 5
same year of NFL play +/- 10

Detailed

Quarterbacks (QBs) who suffered a shoulder injury necessitating placement on the Injured Reserve (IR) list were identified. Placement on the IR indicates a long-term injury rendering players unable to play in the remainder of the season. Pro Sports Transactions IR data from 1980 to 2019 was extracted and entries were filtered for injuries using keywords "shoulder", "labrum", "rotator cuff", “dislocation”, and “AC joint”. An additional manual search of news articles from the NFL, official team websites, and reputable news sources was performed to confirm surgery types and dates and obtain information about players who were placed on the IR without a description of their injury. 65 relevant injuries were found. Of these injuries, 14 were repeat injuries for the same player and 17 were injuries to the non-throwing arm, all of which were excluded. The remaining entries were excluded if the shoulder injury was characterized as a "bruise" or a "strain", and therefore not serious enough injuries to evaluate. Clavicle injuries were also excluded. Players who did not return to play in more than 1 regular season game were excluded for the performance analysis. A total of 19 QBs who received surgery and 11 QBs who suffered a severe shoulder injury but did not receive surgery were included.

QB performance statistics were extracted from Pro Football Reference, which includes statistics by game for players' entire careers. 269 QBs from 1980 to 2020 were found and used as the entire NFL population of QBs. Included performance statistics were selected to be passer rating, passing yards, pass attempts, pass completions, pass completion percentage, passing touchdowns, interceptions, sacks, yards lost to sacks, yards per pass attempt, adjusted yards per pass attempt. Performance statistics were included only if the player attempted more than 1 pass in a game, and statistics were averaged for each game.

Unique age and experience matched controls were selected for each QB who underwent surgery. Controls were matched for experience by finding the set of QBs who started in the NFL at the same age. Using surgery dates and birth dates, a surgery age was calculated for each case QB, which was then used as an index age for possible controls. Pre-injury performance was measured by averaging the Passer Rating per game prior to injury, and finding the control QBs with the closest averaged Passer Rating prior to the index age. Passer Rating was selected as the primary performance statistic due to aggregating other important quarterback statistics and its prevalent usage (NFL Quarterback, NFL Passer Career, Katz, NCAA, Siwoff) in comparing quarterbacks since 1973 (NFL’s, Siwoff). Controls were also selected to have played their first NFL game +/- 10 years from the first NFL game played by the case QB. The list of age, experience, performance, and season matched controls were then selected by those with similar career lengths and 3+ games played previous and post index age. Surgical QB performance vs Surgical Control QB performance was then also visually compared by plotting performance statistics over career length to confirm appropriate controls. The same control matching process was used to find unique controls for QBs who did not receive surgical intervention, with an injury date used as the index, to produce the Non-Surgical QBs and the Non-Surgical Control QBs.

Normality of performance statistics in the entire population of NFL QBs, as well as in the case and control groups were assessed visually with Q-Q and histogram plots, to establish significance in any deviation from the mean in statistical analyses (Morgan). Performance statistics were compared previous and post surgery or injury for the case QBs. Performance statistics were then compared for Surgical and Non-Surgical QBs and their respective controls, prior to surgery, injury, or index age. A gain score analysis (Knapp) on the change in performance pre- and post-injury or index was performed on the case QBs vs control QBs.

Two sample t-tests were performed for each of these sets of statistics, with a paired t-test on the case QBs pre- and post-surgery or injury, and unpaired t-tests for comparisons between case and control QBs. The t-tests were calculated using the Stats package of SciPy 1.6.0 (Virtanen). The null hypothesis for all tests was that the means of compared populations were not significantly different. An alpha of 0.05 was selected, requiring p-values to be below this threshold to reject the null hypothesis. Plots were created using Matplotlib 3.4.2 (Hunter).

Exploratory Data Analysis

Here is the EDA, built using scraped data from publicly available NFL performance metrics and Jupyter Notebook. Data was cleaned and loaded into individual CSV files for every available qualifying quarterback, prior to this step of the EDA.

Below is a table showing the average peformance metric +/- 1 SD for the cohort of quarterbacks who were injured and had surgery and those who were injured and did not have surgery. There is a p-value for each cohort's previous and post performance, as well as the raw numbers of the gain score analysis and p-values associated with those.

Conclusion

Despite small sample sizes, detailed performance metrics of NFL quarterbacks have allowed use to analyze performances with regards to shoulder injury and surgery. We can draw preliminary conclusions from the data that quarterbacks who suffer severe shoulder injuries who are treated surgically, tend to have statistically significant improvements in performances upon their return, when compared to their prior performance, and the performance of quarterbacks who were similarly injured but treated non-surgically.

Further research may include examining greater sample sizes such as collegiate football quarterbacks, a wider range of performance metrics such as pass lengths, and stratification of cohorts based on specific shoulder injury.

There are many limitations to this exploratory analysis. Primarily, the sample size is small, and therefore the power of these statistical tests is smaller. This means that the effects of surgical and non-surgical intervention on quarterback performance should be looked into further, as the data is tending towards surgical intervention improving outcomes but at a very low sample size.

Further, there are numerous outside factors not considered here. Quarterback performances tend to improve with age up to a certain point, as players become more experienced. This could have confounding factors, such as quarterbacks who play more games getting injured more often, better quarterbacks being medically treated differently from other quarterbacks, quarterbacks retiring towards their peaks, etc.

There are also only a few selected statistics used to draw conclusions here, which will no doubt increase in the future as tracking these performance metrics becomes more common and detailed. There will be opportunities in the future to take advantage of these metrics, or even increase the sample size by looking at college football for example.

The data outputted from this exploratory data analysis has been used in the posters and pre-papers of "Performance Following Surgical and Non-surgical Management of Shoulder Injury in NFL Quarterbacks" by Frederick Durrant B.S., George Durrant M.Eng., Martinus Megalla B.A., Teja Makkapati B.A., Nareena Imam B.A., Rocco Bassora M.D., and Frank Alberta M.D., in conjunction with Hackensack Meridian School of Medicine and the Rothman Orthopedic Institute.

References

Enger M, Skjaker SA, Nordsletten L, et al. Sports-related acute shoulder injuries in an urban population. BMJ Open Sport Exerc Med. 2019;5(1):e000551. Published 2019 Aug 12. doi:10.1136/bmjsem-2019-000551

Gibbs DB, Lynch TS, Nuber ED, Nuber GW. Common Shoulder Injuries in American Football Athletes. Curr Sports Med Rep. 2015 Sep-Oct;14(5):413-9. doi: 10.1249/JSR.0000000000000190. PMID: 26359844.

Hunter, J D. Matplotlib: A 2D Graphics Environment. Computing in Science & Engineering. 2007. 9, 90-95.

Kaplan LD, Flanigan DC, Norwig J, Jost P, Bradley J. Am J Sports Med. 2005 Aug; 33(8):1142-6.

Katz S, Burke B. How is Total QBR calculated? We explain our quarterback rating [Internet]. ESPN. 2016 [cited 2021May11]. Available from: https://www.espn.com/blog/statsinfo/post/_/id/123701/how-is-total-qbr-calculated-we-explain-our-quarterback-rating

Kelly BT, Barnes RP, Powell JW, Warren RF. Am J Sports Med. 2004 Mar; 32(2):328-31.

Knapp, Thomas R. and Schafer, William D. (2009) "From Gain Score t to ANCOVA F (and vice versa)," Practical Assessment, Research, and Evaluation: Vol. 14 , Article 6. DOI: https://doi.org/10.7275/yke1-k937

Lanzi JT Jr, Chandler PJ, Cameron KL, Bader JM, Owens BD. Epidemiology of Posterior Glenohumeral Instability in a Young Athletic Population. Am J Sports Med. 2017 Dec;45(14):3315-3321. doi: 10.1177/0363546517725067. Epub 2017 Sep 25. PMID: 28945456.

Monica J, Vredenburgh Z, Korsh J, Gatt C. Acute Shoulder Injuries in Adults. Am Fam Physician. 2016 Jul 15;94(2):119-27. PMID: 27419328.

Morgan, C J. Use of proper statistical techniques for research studies with small samples. Am J Physiol Lung Cell Mol Physiol. Oct 2017. 313, 873-877.

NCAA and NFL Passing Efficiency computation [Internet]. Stassen.com. [cited 2021May11]. Available from: http://football.stassen.com/pass-eff/

NFL Passer Rating Calculator [Internet]. ProFootballReference.com. Sports Reference LLC; [cited 2021May11]. Available from: https://www.pro-football-reference.com/about/qb-rating.htm

NFL Passer Rating Career Leaders [Internet]. Pro-Football-Reference.com. Sports Reference LLC; [cited 2021May11]. Available from: https://www.pro-football-reference.com/leaders/pass_rating_career.html

NFL's Passer Rating [Internet]. Pro Football Hall of Fame Official Site. 2005 [cited 2021May11]. Available from: https://www.profootballhof.com/news/nfl-s-passer-rating/

NFL Quarterback Rating Formula [Internet]. QB Rating Calc - Help. NFL.com; 2011 [cited 2021May11]. Available from: https://web.archive.org/web/20110814052052/http://www.nfl.com/help/quarterbackratingformula

Pallis M, Cameron KL, Svoboda SJ, Owens BD. Epidemiology of acromioclavicular joint injury in young athletes. Am J Sports Med. 2012 Sep;40(9):2072-7. doi: 10.1177/0363546512450162. Epub 2012 Jun 15. PMID: 22707749.

Siwoff, S., & Zimmer, J. (2010). Inside the Numbers. In The Official National Football League Record and Fact Book 2010 (Vol. 91st Season, p. 312). Reference Book, National Football League.

Virtanen P, et al. SciPy 1.0: Fundamental Algorithms for Scientific Computing in Python. Nature Methods. 2020; 17(3), 261-272.