Article, Cardiology

Computerized data mining analysis of keywords as indicators of the concepts in AHA-BLS guideline updates

a b s t r a c t

Introduction: Cardiopulmonary resuscitation (CPR) guidelines have been updated every 5 years since 2000. Sig- nificant changes have been made in each update, and every time a guideline is changed, the instructors of each country that ratify the American Heart Association (AHA) must review the contents of the revised guideline to understand the changes made in the concept of CPR. The purpose of this study was to use a computerized data mining method to identify and characterize the changes in the key concepts of the AHA-Basic Life Support (BLS) updates between 2000 and 2015.

Methods: We analyzed the guidelines of the AHA-BLS provider manual of 2000, 2005, 2010, and 2015 using a computerized data mining method and attempted to identify the changes in keywords along with changes in the guideline.

Results: In particular, the 2000 guideline has focused on the detailed BLS technique of an individual health care provider, whereas the 2005 and 2010 guidelines have focused on changing the ratio of chest compressions and breathing and changing the BLS sequence, respectively. In the most recent 2015 guideline, the CPR team was the central topic. We observed that as the guidelines were updated over the years, keywords related to CPR and automated external defibrillators (AED) associated with co-occurrence network continued to appear.

Conclusions: Analysis revealed that keywords related to CPR and AED associated with the co-occurrence network continued to appear. We believe that the results of this study will ultimately contribute to optimizing AHA’s ed- ucational strategies for health care providers.

(C) 2019

Introduction

The cardiopulmonary resuscitation (CPR) procedure leading to the present one dates to around 1960 [1] when a technical foundation consisting of chest compression [2], artificial respiration [3], and electri- cal defibrillation was established [4]. Since the first CPR guidelines were published by the AdHoc Committee on Cardiopulmonary Resuscitation established by the National Academy of Sciences of the National Re- search Council in 1966, the American Heart Association (AHA) has re- vised the guidelines periodically every 6 years from 1971 to 1992. Furthermore, the CPR guidelines have been updated every 5 years since 2000. However, in the latest scenario, the AHA/International

? The name of organization and date of assembly if the article has been presented: None.

* Corresponding author at: Department of Critical Care and Emergency Medicine, Graduate School of Medicine, University of the Ryukyus, 207 Uehara, Nishihara, Okinawa 903-0215, Japan.

E-mail address: [email protected] (H. Sekiguchi).

Liaison Committee on Resuscitation (ILCOR) process no longer pub- lishes comprehensive guidelines every 5 years. Instead, detailed reviews of the evolving science are performed, and guideline updates are fo- cused only on new science and/or changes that are judged to be important.

Particularly in 2000, the AHA published AHA International Guide- lines 2000 in cooperation with ILCOR. The guidelines were translated not only in the United States but also in other countries, except in En- glish zones, as the first internationally unified guidelines, which have a major impact on countries around the world. In the revised guidelines, ILCOR created the Consensus on Science and Treatment Recommenda- tions (CoSTR) based on a review of the scientific evidence. Furthermore, in the 2010 CoSTR by ILCOR, a new item of Education, Implementation, and Teams was added to improve patient care and outcomes. The EIT states that an effective strategy on methodology is needed. Given this background, if we refer specifically to the AHA-Basic Life Support (BLS) to improve CPR quality and return of spontaneous circulation (ROSC), the 2000 guidelines start with the recommendation of early

https://doi.org/10.1016/j.ajem.2019.11.045

0735-6757/(C) 2019

cardiopulmonary defibrillation using an automated external defibrilla- tor (AED) and a chest compression ventilation ratio of 15:2. In 2005, the guidelines changed the chest compression ventilation ratio from 15:2 to 30:2 and the number of AED shocks. Moreover, the CPR proce- dure was changed from Airway, Breathing, and Circulation (ABC) to Cir- culation, Airway, and Breathing (CAB) in 2010. In the most recent revision in 2015, the concept of team dynamics with an emphasis on chest compression fraction was added. Every time a guideline is changed, the instructors of each country that ratified the AHA must re- view the contents of the revised guideline update to understand the changes made in the concept of CPR presented by the AHA.

Veteran BLS instructors might know from experience the important concept variations that accompany the changes to the guidelines. In- structors might ask, what has AHA pursued to improve CPR quality and ROSC in the historical guidelines transitions? We analyzed the text of the AHA-BLS provider manual (English version) of the 2000, 2005, 2010, and 2015 guidelines by using a computerized data mining method and attempted to identify changes in keywords along with the changes in the guideline. The purpose of this study was to use a computerized data mining method to identify and characterize the changes in the AHA-BLS update key concepts from 2000 to 2015 [5]. We believe that the results of this study will ultimately contribute to op- timizing AHA’s educational strategies for health care providers.

Methods

To identify the changes in the BLS text keywords in the revised AHA guidelines, the following steps were followed to analyze the text data in this study:

The text data in the BLS Provider Manuals (AHA compliant) pub- lished in 2000, 2005, 2010, and 2015 were analyzed. The text of each provider manual (except for tables and figures) was stored in Portable Document Format once and then converted to csv data.
  • A word frequency analysis using the Stanford-POS Tagger [6], which is software that reads text in some language and assigns parts of speech to each word (and other token), such as nouns, verbs, and ad- jectives, was performed. Nouns and proper nouns in the text were analyzed.
  • Co-occurrences between keywords were measured by using Jaccard’s distance. Those co-occurrences were visualized by sub- graph analysis of a co-occurrence network to classify the words into major topics [7]. To determine the edge strength in the co- occurrence network, Jaccard coefficients were calculated. The words with the top 68 were drawn in the diagram, and closely asso- ciated words were color-coded.
  • Correspondence analysis [8] was also used to examine the differ- ences in word usage among the 2000 to 2015 guidelines and to visu- alize them in 2 dimensions. The relative locations between words and groups show the relative frequencies, as in a contingency table. Words that appear on the opposite side of the contrasting group are a characteristic of the target group.
  • A topic model based on latent Dirichlet allocation was also created. In the analysis, 3 latent topics were assumed from the results of the word frequency analysis of each guideline [7]. In addition, to clarify the characteristics of these 3 topics, we deleted words that appeared at the same time on 2 or 3 topics.
  • “R” was used for the analysis regarding #3 to #5 mentioned above [1,9]. This study was not an adaptation of Ryukyu University’s Research Ethics Review for human subjects. Therefore, this study did not require the approval of an ethics committee. Regarding copyright, we obtained a PRINT COPYRIGHT USE AGREEMENT (Inv # 13261-HSEKIGUCHI) from

    the AHA. We deleted the text data stored in a personal computer when the analysis was finally completed according to the AHA regulations.

    Results

    Word frequency analysis

    A total of 104,878 words and 12,044 sentences were extracted from the AHA-BLS health care provider texts from the 2000 to 2015 guidelines. Additionally, we extracted 54,594 words and 4494 lines from the 2000 version, 17,863 words and 2615 lines

    from the 2005 version, 12,092 words and 1745 lines from the 2010 version, and 20,329 words and 3190 lines from the 2015 version. A combined total of 37,580 nouns and proper nouns were used in this keyword analysis. Table 1 shows the top 20 extracted nouns and proper nouns. The most frequently used word in the 2000 to 2015 guidelines was “victim,” followed by “CPR,” “chest,” “AED,” and “compression.”

    Co-occurrence network

    As shown in Fig. 1, co-occurrence between keywords based on the Jaccard’s distance indicated a co-occurrence network diagram of 12 groups. This diagram indicates that there were 12 Wards clusters that co-occurred in the BLS guidelines from 2000 to 2015. Among them are group 1 (G1), related to the airways; G2, related to recognition of car- diac arrest; G3, related to ventilation; G4, related to the rate, cycle, and time of CPR procedures; and G6, related to the AED. In particular, G1, G2, G3, G4, and G6 were connected by a line indicating that they were strongly related in the 2000 to 2015 guidelines. These clusters re- lated to the CPR and AED technique’s words formed a co-occurrence network and were relevant in all 4 changes to the guidelines. In other words, words related to CPR and AED were considered important BLS concepts that appeared consistently regardless of changes in the guide- lines. There were also 7 clusters that became key concepts of BLS in the 4 guideline updates: “Chain of survival” (G5), “Adult, child, and infant age categories” (G7), “Team resuscitation” (G8), “Healthcare Provider” (G9), “Defibrillation and survival” (G10), “Emergency response system and emergency response number” (G11), and “BLS sequence” (G12). Each cluster is a theme that has been covered in the guidelines and can be thought of as keywords related to important concepts of BLS common to the 2000 to 2015 guidelines extracted in the word fre- quency analysis.

    Table 1 The top 20 words (nouns and proper nouns only) based on word frequency analysis of the 2000-2015 guidelines.

    Words

    Frequency

    1

    victim

    1533

    2

    CPR

    886

    3

    chest

    832

    4

    AED

    737

    5

    compression

    704

    6

    rescuer

    704

    7

    breath

    629

    8

    infant

    581

    9

    arrest

    544

    10

    breathing

    506

    11

    airway

    462

    12

    mask

    411

    13

    emergency

    406

    14

    rescue

    349

    15

    sign

    344

    16

    pulse

    342

    17

    response

    327

    18

    shock

    316

    19

    head

    296

    20

    pad

    295

    CPR; cardiopulmonary resuscitation, AED; automated external defibrillator.

    Fig. 1. Co-occurrence network diagram of the AHA Guidelines (2000-2015). The co- occurrence between keywords was based on the Jaccard’s distance. Group 1 (G1) is related to the airway; G2, recognition of cardiac arrest; G3, ventilation; G4, rate, cycle, and time of CPR procedures; G6, AED; G7, adult, child, and infant age categories; G8, team resuscitation; G9, Health care provider; G10, defibrillation and survival; G11, emergency response system; and G12, emergency response number related to the BLS sequence.

    Correspondence analysis

    The results of the correspondence analysis are shown in Fig. 2. Examples of words that characterized Guideline 2000-BLS were foreign-body airway obstruction (FBAO), number, circulation, in- jury, rescue, cause, neck, sign, AED oxygen, and breathing; exam- ples that characterized Guideline 2005-BLS were cycle, rate, breath, chin, and tilt; examples that characterized 2010-BLS were rescuer, step, seconds, sequence, compression, and ventilation; and examples that characterized 2015-BLS were team, life, BLS, adult, and resuscitation.

    Topic model

    The 3 latent topics that were analyzed from the results of the word frequency analysis of each guideline are shown in Table 2. Fifty words were automatically extracted for each topic using a computer. Next, within those topics, words that appeared at the same time in 2 or 3 topics were deleted, and the remaining words in each topic are shown in Table 3. Ultimately, 12 words, namely, AED, air, cause, circulation, de- fibrillation, FBAO, healthcare, hospital, injury, neck, number, and oxy- gen, remained as characteristic words for Topic 1; 9 words, namely, blood, chin, cycle, lift, rate, rescuer, sequence, thumb, and tilt, remained as characteristic words for Topic 2; and 6 words, namely, BLS, body, life, side, survival, and team, remained as characteristic words for Topic 3. Furthermore, the percentage of each topic in each guideline is shown vi- sually in Fig. 3. In Fig. 3, Topic 1 is in red, Topic 2 is in blue, and Topic 3 is in green. Topic 1 accounted for approximately 80% of the 2000 guide- line, and Topic 2 accounted for approximately 90% of the 2005 guideline and approximately 70% of the 2010 guideline. Topic 3 increased (ap- proximately 20%) in the 2010 guideline and increased to approximately 60% of the 2015 guideline.

    Discussion

    Historically, in pursuit of higher ROSC, AHA CPR guidelines have pur- sued high-quality CPR based on scientific evidence. Every time the guidelines are updated, BLS instructors from global countries that have ratified the AHA must review the key concepts of the new guide- lines and then teach them in their BLS classes. However, the changes in the BLS key concepts do not appear to incorporate all of the concepts reflected by the analysis of the AHA-BLS provider manual. This is the first study that has compared the changes in BLS key concepts with the results of data mining analysis of the BLS provider manual officially approved by the AHA.

    We would like to emphasize the following 3 observations shown by the results obtained in this study. First, regarding the co-occurrence net- work of BLS key concept words, despite the 4 revisions to the guidelines, the strong associations among G1, G2, G3, G4, and G6 related to CPR and AED indicated a co-occurrence network of those groups. In other words, words related to CPR and AED are considered important BLS concepts

    Fig. 2. Plot of the correspondence analysis of the guidelines. The Euclidean distance among the guidelines is a measure of the similarity of their word distribution. The top 60 words filtered by their chi-square values were plotted. These words are representative of the closest guidelines. The vertical axis and horizontal axis in the figure each indicate that the percentage con- tribution of the principal components using 2 dimensions is 71.7% + 22.4% = 94.1%.

    Table 2

    Three latent topics analyzed from the results of the word frequency analysis of each guideline.

    Topic 1 Topic 2 Topic 3

    action, AED, air, airway arrest, bag, breath, breathing, care, cause, chest, child, circulation, compression, CPR, defibrillation, device, emergency, face, FBAO, finger, hand, head, healthcare, heart, hospital infant, injury, jaw, mask, minute, mouth, neck, number, oxygen, position, provider, rescue, rescuer, response, resuscitation, rhythm, shock, sign, thrust, time, use, victim

    year

    adult, AED, airway, arrest, bag, blood, breath, breathing, chest, child, chin, compression, CPR, cycle, emergency, face, finger, hand, head, heart infant, jaw, lift, mask, minute, mouth, obstruction

    pad, position, provider, pulse, rate, rescuer, response, rhythm, seconds, sequence, shock, sign, step, system, technique, thumb, tilt, time, use

    ventilation, victim, year

    AED, action, adult, airway, arrest, bag, BLS, body, breath, breathing, care, chest, child, compression, CPR, device emergency, finger, hand, head, heart, infant, jaw, life, mask, minute, mouth, obstruction, pad, provider, pulse, rescue, rescuer

    response, resuscitation, rhythm, seconds, shock, side, sign, step, survival system, team, technique, thrust, use, ventilation, victim, year

    Fig. 3. The topic models of each guideline. Topic 1 is in red, Topic 2 is in blue, and Topic 3 is in green. In the 2000 guideline, Topic 1 accounted for approximately 80%. In the 2005 guideline, Topic 2 accounted for approximately 90%. In the 2010 guideline, Topic 2 also accounted for approximately 70%, and Topic 3 increased (approximately 20%). In the 2015 guideline, Topic 3 increased to approximately 60% of the total. The contents of each topic are shown in Table 3.

    AED; automated external defibrillator, CPR; cardiopulmonary resuscitation, FBAO; foreign body airway obstruction, BLS; basic life support.

    that appear consistently regardless of changes in the guidelines. Second, regarding the characteristic words of each guideline based on corre- spondence analysis, the frequency of FBAO indicated that it was a major characteristic word in the 2000 guideline. We believe that the characteristics of the 2000 guideline are well expressed by the words “protection of neck in injury,” “circulation sign,” “rescue breathing,” “use of oxygen,” and “use of AED.” The characteristics of the 2005 guide- line were indicated by “cycle” and “rate,” which were related to the change in the chest compression and ventilation ratio from 15:2 to 30:2 in one cycle, and by “chin-tilt” and “breath,” which were related to breathing after the airway is secured. Compared with the words in the other guidelines, “rescuer” was the most frequently used and dis- tinctive word in the 2010 guideline. Furthermore, we found that “com- pression,” “ventilation,” “step,” and “sequence” were related to the change from ABC to CAB in the BLS procedure. The most characteristic word of the 2015 guideline was “team,” which was related to “BLS on team” or “team dynamics.” In addition, we assumed that the word “life” reflected the item “life is why,” which is the slogan set by AHA as the reason for our actions to promote the spread of cardiopulmonary resuscitation. Third, the transition of feature words on the basis of topic analysis showed that AED, defibrillation, and response to FBAO indi- cated that the BLS technique was the main topic in the 2000 guideline. In the 2005 guideline, the word “rate” was frequently used and was as- sociated with the change in the rate of chest compression/ventilation from 15:2 to 30:2 and the change in the AED shock rate from 3 shocks to 1 shock. In the 2010 guideline, ABC was changed to CAB in the

    Table 3

    Words specific to each topic.

    Topic 1 Topic 2 Topic 3

    AED blood BLS

    air chin body

    cause cycle life

    circulation lift side

    defibrillation rate survival

    FBAO rescuer team

    health care sequence

    hospital thumb

    injury tilt

    neck number oxygen

    AED; automated external defibrillator, FBAO; foreign body airway obstruction, BLS; basic life support.

    sequence of CPR, and “chin-lift,” “chin-tilt,” “thumb,” “cycle,” and “se- quence” were considered to be related to the deletion of the confirma- tion of breathing with chin-tilt “look, listen, feel.” The use of the words “BLS” and “team” indicated an increasing trend in Topic 3 from the 2005 guideline because BLS was recommended to be conducted by mul- tiple rescuers rather than by a single rescuer. Additionally, the words “BLS” and “team” in Topic 3 were thought to be important messages of CPR performed by a team and related to team dynamics in the 2015 guidelines. The 2000 guideline emphasized detailed techniques for indi- vidual BLS, but to improve the Quality of CPR, a major technique change was made in the 2005 and 2010 guidelines. The 2015 guideline focused on the importance of performing BLS by a team. The analysis indicated a trend of BLS performed by individuals to BLS performed by a team of multiple rescuers.

    A survey for finding keywords that exist in a large amount of text data is called “text mining,” which is “data mining” for text. This re- search method is included in the category of “content analysis.” Content analysis is an investigation method that makes inferences by systemat- ically and objectively identifying specific features in text [10]. Previous qualitative studies may have relied on human work to count the fre- quency of certain words. However, this approach required a huge amount of human effort and time, and the work process was compli- cated. However, by using a computer to classify morphemes in text and extract keywords, we were able to significantly reduce work pro- cess time. In addition, computerization has made the reproducibility of work contents clear and has been used for research in many fields [11,12]. Text analysis is now widely used in many academic fields, and a quantitative approach using a computer is also useful in qualitative research.

    This study performed data mining of English texts, but by using the methods in this study, it should be possible, for example, to verify the equivalence in the types and frequencies of words that appear as impor- tant concepts between English guidelines and those translated into other languages. Such a validation would assess whether the guideline concept is faithfully communicated throughout the world. We may also be able to identify differences in keywords between the European Resuscitation Council and the AHA as well as the differences in concepts between other resuscitation organizations. We are planning to compare the contents of the same BLS guidelines in English and Japanese as the next stage of research.

    There were a few limitations in this study. First, the keywords of this study were recognized by higher frequency of usage. This high occur- rence frequency does not necessarily reflect the importance of words, and certain important keywords may appear infrequently. Second, the analysis was limited to nouns and proper nouns. If “verb” had been in- cluded in the analysis, some differences in the analysis results from

    those in this study could have occurred. Finally, although our research method was computerized data mining, the interpretation and discus- sion of the results after keyword extraction were performed by AHA- BLS instructors, so it was not a completely automated analysis.

    Conclusions

    We performed computerized data mining for the AHA-BLS guide- lines. The analysis showed that, as the guidelines were updated over the years, keywords related to CPR and AED associated with the co- occurrence network continued to appear. In particular, the 2000 guide- line focused on the detailed BLS technique of an individual health care provider. The 2005 and 2010 guidelines focused on changing the ratio of chest compressions and breathing and changing the BLS sequence, re- spectively. In the most recent 2015 guideline, team-BLS was the central topic.

    Funding sources/disclosures

    There was no specific funding for this study.

    Author contributions section Hiroshi Sekiguchi: Conceptualization, Methodology, Data Curation,

    Writing – Original Draft, Writing – Review & Editing Tatsuma Fukuda: Supervision Yuichiro Tamaki: Investigation Kazuhiko Hanashiro: Pro- ject administration Kenichi Satoh: Software, Formal analysis, Visualiza- tion Eiichi Ueno: Methodology, Investigation Ichiro Kukita: Data Curation, Validation.

    Acknowledgments

    We thank the AHA (Dallas, US) for approving the PRINT COPYRIGHT USE AGREEMENT in the AHA-BLS provider manuals that enabled us to

    achieve our project purpose of comparing keywords and the resuscita- tion concepts from the 2000, 2005, 2010, and 2015 guidelines updates.

    References

    1. DeBard ML. The history of cardiopulmonary resuscitation. Ann Emerg Med 1980;9: 273-5. https://doi.org/10.1016/s0196-0644(80)80389-1.
    2. Kouwenhoven WB, Jude JR, Knickerbocker GG. Closed-chest cardiac massage. JAMA 1960;173. https://doi.org/10.1001/jama.1960.03020280004002 1064-17.
    3. Safar P, Escarraga LA, Elam JO. A comparison of the mouth-to-mouth and mouth-to- airway methods of artificial respiration with the chest-pressure arm-lift methods. N Engl J Med 1958;258:671-7. https://doi.org/10.1056/NEJM195804032581401.
    4. Zoll PM, Linenthal AJ, Gibson W, Paul MH, Norman LR. Termination of ventricular fi- brillation in man by externally applied electric countershock. N Engl J Med 1956; 254:727-32. https://doi.org/10.1056/NEJM195604192541601.
    5. Stone PJ. The general inquirer: a computer approach to Content Analysis. Cam- bridge: MIT Press; 1966.
    6. Tagger S-P. Avialable at https://nlp.stanford.edu/software/tagger.shtml.
    7. Ghorbani F, Feizabadi M, Farzanegan R, Vaziri E, Samani S, Lajevardi S, et al. An in- vestigation of topics and trends of tracheal replacement studies using co- occurrence analysis. Tissue Eng B 2017;23:118-27. https://doi.org/10.1089/ten. TEB.2016.0254.
    8. Fajardo-Ortiz D. Ortega-Sanchez-de-Tagle J, Castano VM. Hegemonic structure of basic, clinical and patented knowledge on Ebola research: a US Army reductionist initiative. J Transl Med 2015;13:124. https://doi.org/10.1186/s12967-015-0496-y.
    9. R Core Team. R: a language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2018 Available from: https:// www.R-project.org/ [November 14, 2019].
    10. Toutanova K, Klein D, Manning C, Singer Y. Feature-rich part-of-speech tagging with a cyclic dependency network. Proceedings of HLT-NAACL 2003; 2003. p. 252-9.
    11. Matsuo M, Kawano N, Satoh K, Apsalikov KN, Moldagaliev T. Overall image of nu- clear tests and their human effects at semiparatincs an attempt at analyses based on verbal data. J Radiat Res 2006;47:A219-24. https://doi.org/10.1269/jrr.47.a219.
    12. Hasegawa A, Koike M, Nemoto M, Ohba T, Yamada C, Matsui S, et al. Lexical analysis suggests differences between subgroups in anxieties over radiation exposure in Fukushima. J Radiat Res 2018;59:ii83-90. https://doi.org/10.1093/jrr/rry027.

    Leave a Reply

    Your email address will not be published. Required fields are marked *