Surgery

A new approach to the management of acute appendicitis: Decision tree method

a b s t r a c t

Background: It is important to distinguish between complicated acute appendicitis (CAA) and noncomplicated acute appendicitis (NCAA) because the treatment methods are different. We aimed to create an algorithm that determines the severity of Acute appendicitis without the need for imaging methods, using the decision tree method.

Methods: The patients were analyzed retrospectively and divided into two groups as CAA and NCAA. Age, gender, Alvarado scores, white blood cell values (WBC), neutrophil/lymphocyte ratios (NLR), C-reactive protein value (CRP), albumin value and CRP/Albumin ratios of the patients were recorded.

Results: In the algorithm we created, the most important parameter in the distinction between CAA and NCAA is CRP. NLR is predictive in patients with a CRP value of <=107.565 mg/L, and the critical value is NLR 2.165. In pa- tients with a CRP value of >107.565 mg/L, albumin is the determinant and the critical value is 2.85 g/dL. Age, gen- der, alvarado score and CRP/albumin ratio have no significance in distinguishing between CAA and NCAA. In the statistical analysis, there were significant differences between NCAA and CAA groups in terms of age (39.56 years vs 13,675 years), gender (48.1% male vs 71.4% male), WBC (13,891.10/mL vs 11,614.76/mL), CRP (27 mg/L vs 127 mg/L), albumin (3 g/dL vs 3 g/dL) and CRP/albumin (9.50 vs. 41).

Conclusion: Thanks to the algorithm we created, CAA and NCAA distinction can be made quickly. In addition, by avoiding unnecessary surgical procedures in NCAA cases, patients’ quality of life can be increased and morbidity rates can be minimized.

(C) 2022

  1. Introduction

acute appendicitis accounts for approximately 3.8% of all emer- gency department admissions due to abdominal pain; and appropriate therapy should be started as soon as its diagnosis is made [1]. Although appendectomy is the gold standard treatment for acute appendicitis, some studies advocate that non-complicated acute appendicitis (NCAA) cases can be treated non-surgically [2,3]. Those studies have shown that antibiotic treatment does not increase the risk of perforation in NCAA. Additionally, considering that the number of surgical proce- dures and hospital admissions would be reduced, one can argue that

* Corresponding author at: Baskent University, Department of General Surgery, Yukari Bahcelievler, Maresal Fevzi Cakmak St. No: 45, 06490 Cankaya, Ankara, Turkey.

E-mail addresses: [email protected] (E. Karakaya), [email protected] (S.C. Yucebas).

1 Postal address: Baskent University, Department of General Surgery, Yukari Bahcelievler, Maresal Fevzi Cakmak St. No: 45, 06490 Cankaya, Ankara, Turkey

2 Postal address: Canakkale Onsekiz Mart Univesity, Faculty of Engineering, Computer Engineering Department, arbaros, 17,100 Kepez/Canakkale

antibiotic treatment is an advantageous treatment method in NCAA cases in terms of both patient’s quality of life and treatment cost.

There are various appendicitis scoring systems in the literature to di- agnose acute appendicitis, such as Alvarado, appendicitis inflammatory response, and adult appendicitis score [4]. These scores usually use clin- ical and laboratory parameters. They are highly inadequate to make a distinction between CAA and NCAA, a task which typically requires radiological studies [5].

It is important to distinguish complicated acute appendicitis (CAA) and NCAA in the diagnostic process to select the appropriate treatment. The European Association of Endoscopic Surgery defines CAA as gangre- nous AA with or without by perforation, AA with intraabdominal ab- scess, and AA with periappendicular phlegmon or purulent/free fluid [6]. While scoring systems are unable to distinguish between compli- cate acute appendicitis and NCAA, imaging studies can help make this distinction only in selected cases [7,8].

In this study, we aimed to create an algorithm to distinguish between CAA and NCAA with a decision tree method using laboratory parameters alone in patients with high suspicion for AA. Thanks to

https://doi.org/10.1016/j.ajem.2022.01.034

0735-6757/(C) 2022

this algorithm we will create, CAA and NCAA cases will be rapidly and easily distinguished, and additional tests such as imaging studies, which are costly and time-consuming, will not be needed. As a result,

A: Attribute A = {a1, a2, …, av.} In this case, the information needed to classify the samples is calcu-

lated as:

this algorithm will mediate the selection of the most effective treatment

? (S , S , … S )=

m

pi log 2 pi

in terms of both patient’s quality of life and cost.

1 2 m

–? .

i=1

  1. Methods

The entropy measure for attribute A that will divide the sample set into v subsets is calculated as follows:

Our study was approved by Baskent University Clinical Research Ethics Committee at 09/03/2021 with the study number KA21/109. A total of 872 patients, who were operated with appendectomy between January 2011 and December 2020 at our hospital, were retrospectively

EA =

X S1i + S2i + … + Smi I (S1i i=1 S

, …, Smi)

k

analyzed via hospital automation system. One hundred and twenty- one patients were excluded, who were found to have a Malignant disease or not to have appendicitis by the pathology examination. The patients were grouped in two groups as CAA and NCAA. The CAA group (n = 42) included gangrenous AA, AA with intraAbdominal abscess, and AA with periappendicular phlegmon or purulent/free fluid while the NCAA group included the other patients (n = 709) (Fig. 1). The demographic characteristics (age, gender), laboratory results (white blood cell (WBC), neutrophil lymphocyte ratio (NLR), C-reactive protein (CRP), albumin, CRP albumin ratio (CAR)), physical examination findings, and Alvarado score of the patients were recorded and analyzed.

    1. Desicion tree method

Decision trees are widely used in machine learning applications in the field of medicine with its easy interpretation and effective perfor- mance in the solution of nonlinear problems [9]. A decision tree model uses various statistical calculations in order to place the given attribute within the nodes of the tree structure [10]. These calculations are aimed to find the strongest parameter that can homogeneously classify the given dataset. Each available parameter is tested for the relevant node, and the strongest parameter is selected. The branches connecting the node to the subnodes express the test condition according to the values that the current node can take. In order to calculate the strongest attri- bute for a given node, entropy based information gain ratio is used as in the study reported by Sancak et al. [11]. Suppose that:

Si: Samples of class Ci

Pi: Probability of data i belongs to class Ci

Image of Fig. 1

Fig. 1. Flowchart of the study population (CAA: complicated acute appendicitis; NCAA: non-complicated acute appendicitis; AA: acute appendicitis).

Sij indicates the samples in class Ci that belongs to subset Sj, the

information for subset Sj is calculated as:

I (S1, S2, …Sm) = — X pi log2 pi

m

i=1

In this case, information gain is the difference between information and entropy and calculated as given below:

GAINA = I (S1, S2, … Sm) — EA

The maximum depth for the C4.5 decision tree is set to 10, minimum leaf size is set to 2, and minimum size of split is given as 4. The perfor- mance of the decision model is calculated by using 10-fold cross valida- tion. Examples in each fold is chosen randomly. The binary class ratio of the original dataset is also kept in each fold to avoid bias. The accuracy of the model is calculated as 95.48%.

    1. Statistical analysis

For categorical variables, mean +- standard deviation or median (minimum-maximum) was used for number and percentage numerical variables, depending on the data distribution. The normal distribution of data was evaluated using the Shapiro Wilks test. Comparison of numer- ical measurements according to socio-demographic characteristics and research groups Mann Whitney U test and Student t-test was used for two independent groups in accordance with the data distribution. Pro- portion comparisons or correlation studies according to research groups were investigated using the Chi-square or Fisher’s exact test. For the sta- tistical significance level, p < 0.05 was accepted.

  1. Results

When we analyzed the patient data by the decision tree method, we determined that CRP was the most important parameter for distin- guishing between CAA and NCAA. In patients with a CRP level of

107.565 mg/L or less, albumin is the first parameter that should be evaluated. While NLR value is important in patients with an Albumin level of 2.85 g/dL or less, WBC is important when albumin level is above 2.85 g/dL. Patients with a NLR value of 1.6 or less have NCAA with a likelihood of 99%, and those with a NLR value below 1.6 have 100% CAA. In patients with an albumin level above 2.85 g/dL, those with a WBC count of 9200/mL or less are 100% likely to have NCAA, and those with a WBC count of 9200/mL are 70% likely to have CAA.

In patients with a CRP level above 107.565 mg/dL, NLR is the first pa- rameter that should be evaluated. The patients with a NLR level of 2.165 or less have NCAA with a likelihood of 69.5%, albumin should be evalu- ated for those with a NLR level above 2.165. While the patients with an albumin level of 3.25 g/dL are 81% likely to have CAA, patients with an albumin level above 3.25 g/dL should have their WBC evaluated. While patients with a WBC count of 9450/mL or below are 96% likely to have NCAA, those with a WBC count above that number are 73% likely to have CAA (Fig. 2).

Image of Fig. 2

Fig. 2. Algorithm developed with decision tree to detect complicated appendicitis (CRP: C- reactive protein; NLR: neutrophil lymphocyte ratio; WBC: White blood cell; CAA: complicated acute appendicitis; NCAA: non-complicated acute appendicitis).

A statistical analysis of the patients showed that there was no signif- icant difference between the study groups regarding sex, Alvarado score, and NLR whereas the two groups differed significantly with respect to age, WBC, CRP, Albumin, and CAR (Table 1).

Table 1

Comparing demographic data and laboratory values between groups. (WBC: white blood cell; NLR: neutrophil lymphocyte ratio; CRP: C-reactive protein).

NCAA (n:709)

CAA (n:42)

p value

Age

39.56 +- 13.675

49.48 +- 17.497

0.000?

Gender Male

341 (48.1%)

30 (71.4%)

0.003?

Female

368 (51.9%)

12 (28.6%)

Alvarado score

5.09 +- 1.625

5.10 +- 1.875

0.705?

WBC

13,891.10 +- 9290.107

11,614.76 +- 4452.062

0.008?

NLR

6 (1-101)

5.50 (1-23)

0.637+

CRP

27 (1-300)

127 (2-300)

<0.001+

Albumin

3 (2-5)

3 (2-4)

<0.001+

CRP/Albumin

9.50 (1-118)

41 (1-76)

<0.001+

+ Mann Whitney U Test (median (min-max)).

* Student-t-Test (mean +- sd).

? Chi-square test.

  1. Discussion

According to the decision tree algorithm that we created in our study, CRP is the most important parameter for distinguishing CAA and NCAA in patients with AA; however, age, sex, Alvarado score, and CRP/albumin ratio had no effect on this distinction.

Studies in the literature have emphasized that the incidence of AA is highest between the ages 10 and 30 years and decreases with Advancing age [12-14]. Those studies have emphasized that the reduced number of lymphoid follicles in appendix vermiformis with advancing age causes a decrease in the incidence of AA. On the other hand, it has been reported that the incidence of AA increases in older patients due to a longer life ex- pectancy especially in developed and Developing countries [15]. While the Morbidity and mortality rates of AA are 1% in the general population, they may increase up to 70% due to a higher risk of CAA in the elderly [16]. Furthermore, many studies have reported that AA is more common in men [17-19]. On the other hand, Ashley et al. reported that CAA is more common in women [20]. Although our statistical analysis revealed signif- icantly higher percentages of the elderly and men in the CCA group than the NCAA group, age and sex had no effect on the distinction between CCA and NCAA in our algorithm. We believe that this likely resulted

from the fact that the mean age of the study population was below 60 years and both groups had similar sex distributions. More accurate results can be obtained by a future study with a patient population including more AA patients that are older than 60 years of age. Still, considering that the incidence of AA is lower in the elderly than in the general popu- lation, an effective treatment can be planned using our algorithm.

Alvarado score is a clinical scoring system that has been used to di- agnose AA since 1986. This scoring system consists of 8 parameters: lo- calized tenderness in the right lower quadrant, leucocytosis, migration of pain, leukocytosis with left shift, fever, nausea-vomiting, anorexia, and rebound pain [21]. Although Alvarado scoring system is not suffi- cient for making the diagnosis, a score of less than 5 has a sensitivity of 99% for the exclusion of AA [4]. In accordance with the literature, our study also showed that the Alvarado score does not make any con- tribution to the distinction between CAA and NCAA, both in our algo- rithm and the statistical analysis.

Elevated WBC count has been shown by many studies as an impor- tant laboratory parameter making the diagnosis of AA [22,23]. Guraya et al. reported that elevated leucocytes can be used not only to diagnose AA, but also to determine its severity [24]. NLR is a laboratory parameter reflecting subclinical inflammation, which can be readily calculated using differential leucocyte levels [25]. While WBC level is a parameter that shows active and ongoing inflammation, lymphocyte level is related to the regulatory pathway [26]. Many studies using this mechanism have emphasized that NLR is a reliable marker of the severity of acute appen- dicitis [27]. CRP, a positive acute phase protein, is elevated in inflamma- tory conditions. Xharra et al. reported that CRP had a positive predictive value of 94.7%, a specificity of 72%, and a sensitivity of 85.1% for acute ap- pendicitis [28]. In another study, Worm et al. demonstrated that CRP elevation can be used to distinguish between CAA and NCAA [29]. CRP is known as positive acute phase protein, and albumin is known as neg- ative acute phase protein [30]. Therefore, we think that the CAR will de- tect inflammation more clearly. There is a limited number of studies in the related literature. Dogan S. et al. showed in their study that if the CAR is >4.4 in acute appendicitis, CA may be present [31].

While CRP is the most important parameter for distinguishing be- tween CAA and NCAA in our algorithm, the parameters albumin, NLR, WBC, and albumin also have a strong effect on this distinction. Although a significant difference was found between the study groups regarding CAR, it was not present in our algorithm.

The limitations of our study are its single center design and a lower number of patients in the CAA group. A multi-center prospective study with a larger population may include a greater number of param- eters and design a more advanced algorithm.

As a conclusion, AA is a condition which should be appropriately treated once it is diagnosed. NCAA cases can be usually successfully treated with non-surgical methods. Our algorithm allows fast and easy detection of NCAA cases, in situations where radiological imaging re- sources are limited and computed tomography is not appropriate, such as pregnancy. By this way, AA patients can be treated successfully in the most cost-effective and least morbid manner by avoiding unnec- essary surgical procedures.

Authorship

Made a substantial contribution to the concept or design of the work; or acquisition, analysis or interpretation of data: Murathan ERKENT, Sait Can YUCEBAS

Drafted the article or revised it critically for important intellectual content: Emre KARAKAYA

Approved the version to be published: Emre KARAKAYA

Ethics committee approval

The study protocol was approved by the Baskent University Ethics Committee for Clinical Research (project no. KA21/109)

Financial disclosure

The authors of this manuscript have no conflicts of interest to disclose.

Declaration of Competing Interest

We have no conflicts of interest to disclose.

References

  1. Cervellin G, Mora R, Ticinesi A, Meschi T, Comelli I, Catena F, et al. Epidemiology and outcomes of acute abdominal pain in a large urban emergency department: retro- spective analysis of 5,340 cases. Ann Transl Med. 2016;19:362.
  2. Podda M, Gerardi C, Cillara N, Fearnhead N, Gomes CA, Birindelli A, et al. Anti- biotic treatment and appendectomy for uncomplicated acute appendicitis in adults and children: a systematic review and Meta-analysis. Ann Surg. 2019; 270:1028-40.
  3. Collaborative C, Flum DR, Davidson GH, Monsell SE, Shapiro NI, Odom SR, et al. A randomized trial comparing antibiotics with appendectomy for appendicitis. N Engl J Med. 2020;20:1907-19.
  4. Podda M, Pisanu A, Sartelli M, Coccolini F, Damaskos D, Augustin G, et al. Diagnosis of acute appendicitis based on Clinical scores: is it a Myth or reality? Acta Biomed. 2021.;92:e2021231.
  5. Hoffmann JC, Trimborn CP, Hoffmann M, Schroder R, Forster S, Dirks K, et al. Classi- fication of acute appendicitis (CAA): treatment directed new classification based on imaging (ultrasound, computed tomography) and pathology. Int J Colorectal Dis. 2021;36:2347-60.
  6. Gorter RR, Eker HH, Gorter-Stam MA, Abis GS, Acharya A, Ankersmit M, et al. Diag- nosis and Management of acute appendicitis. EAES consensus development confer- ence 2015. Surg Endosc. 2016;30:4668-90.
  7. Giljaca V, Nadarevic T, Poropat G, Nadarevic VS, Stimac D. Diagnostic accuracy of ab- dominal ultrasound for diagnosis of acute appendicitis: systematic review and Meta- analysis. World J Surg. 2017;41:693-700.
  8. Krajewski S, Brown J, Phang PT, Raval M, Brown CJ. Impact of computed tomography of the abdomen on clinical outcomes in patients with acute right lower quadrant pain: a meta-analysis. Can J Surg. 2011;54(1):43-53.
  9. Sharma H, Kumar S. A survey on decision tree algorithms of classification in data mining. Int J Sci Res. 2016;5:2094-7.
  10. Kokol P, Mernik M, Zavrsnik J, Kancler K, Malcic I. Decision trees based on automatic learning and their use in cardiology. J Med Syst. 1994;18:201-6.
  11. Sancak EB, Kilinc MF, Yucebas SC. Evaluation with decision trees of efficacy and safety of Semirigid Ureteroscopy in the treatment of proximal ureteral calculi. Urol Int. 2017;99:320-5.
  12. Rahman MA, Parash TH, Banu LA. Diameter of the lymphoid follicles in the vermiform appendix of Bangladeshi cadaver. Mymensingh Med J. 2014;23: 315-20.
  13. Avanesov M, Wiese NJ, Karul M, Guerreiro H, Keller S, Busch P, et al. Diagnostic pre- diction of complicated appendicitis by combined clinical and radiological appendici- tis severity index (APSI). Eur Radiol. 2018;28:3601-10.
  14. Yamini D, Vargas H, Bongard F, Klein S, Stamos MJ. Perforated appendicitis: is it truly a surgical urgency? Am Surg. 1998;10:970-5.
  15. Storm-Dickerson TL, Horattas MC. What have we learned over the past 20 years about appendicitis in the elderly? Am J Surg. 2003;185:198-201.
  16. Horattas MC, Guyton DP, Wu D. A reappraisal of appendicitis in the elderly. Am J Surg. 1990;160:291-3.
  17. Lee JA. The influence of sex and age on appendicitis in children and young adults. Gut. 1962;3:80-4.
  18. Madiba TE, Haffejee AA, Mbete DL, Chaithram H, John J. Appendicitis among African patients at King Edward VIII Hospital, Durban, South Africa: a review. East Afr Med J. 1998;75:81-4.
  19. Rahman MM, Begum J, Khalil M, Latif SA, Nessa A, Jahan MK, et al. Histomorphological study of lymphoid follicle of vermiform appendix. Mymensingh Med J. 2008;17:134-40.
  20. Ashley DJ. Observations on the epidemiology of appendicitis. Gut. 1967;8:533-8.
  21. Alvarado A. A practical score for the early diagnosis of acute appendicitis. Ann Emerg

Med. 1986;15:557-64.

  1. Andersson RE. Meta-analysis of the clinical and laboratory diagnosis of appendicitis. Br J Surg. 2004;91:28-37.
  2. Kharbanda AB, Taylor GA, Fishman SJ, Bachur RG. A clinical decision rule to identify children at low risk for appendicitis. Pediatrics. 2005;116:709-16.
  3. Guraya SY, Al-Tuwaijri TA, Khairy GA, Murshid KR. Validity of leukocyte count to predict the severity of acute appendicitis. Saudi Med J. 2005;26:1945-7.
  4. Kaser SA, Fankhauser G, Willi N, Maurer CA. C-reactive protein is superior to biliru- bin for anticipation of perforation in acute appendicitis. Scand J Gastroenterol. 2010; 45:885-92.
  5. Acarturk G, Acay A, Demir K, Ulu MS, Ahsen A, Yuksel S. Neutrophil-to-lymphocyte ratio in inflammatory bowel disease – as a new predictor of disease severity. Bratisl Lek Listy. 2015;116:213-7.
  6. Hajibandeh S, Hajibandeh S, Hobbs N, Mansour M. Neutrophil-to-lymphocyte ratio predicts acute appendicitis and distinguishes between complicated and uncompli- cated appendicitis: a systematic review and meta-analysis. Am J Surg. 2020;219: 154-63.
  7. Xharra S, Gashi-Luci L, Xharra K, Veselaj F, Bicaj B, Sada F, et al. Correlation of serum C-reactive protein, white blood count and neutrophil percentage with histopathology findings in acute appendicitis. World J Emerg Surg. 2012;7:27.
  8. Van den Worm L, Georgiou E, De Klerk M. C-reactive protein as a predictor of sever- ity of appendicitis. S Afr J Surg. 2017;55:14-7.
  9. Roberts I, Blackhall K, Alderson P, Bunn F, Schierhout G. Human albumin solution for resuscitation and Volume expansion in critically ill patients. Cochrane Database Syst Rev. 2011.;11:CD001208.
  10. Dogan S, Kalafat UM, Bildik B, Sarici IS, Cander B. Diagnostic value of C-reactive pro- tein/albumin ratio to differentiate simple versus complicated appendicitis. Eur J Emerg Med. 2020;19:178.