A multiple logistic model for prediction of urinary tract infections in an urban community: A public health perspective

Introduction

Urinary tract infection (UTI) is one of the most common infectious diseases seen in ambulatory patients, with a 50-60% prevalence rate in adult females.1 It imposes a significant socioeconomic burden, leading to an increase in hospital visits, absenteeism, and a lower quality of life. Antibiotic resistance among uropathogens, particularly those causing community-acquired UTIs, has recently emerged as one of the most challenging public health concerns and significant attributes influencing UTI care and management.2 Community-acquired UTI is the second most encountered microbial infection in the community setting.3 Urinalysis and urine culture with antibiotic susceptibility are used to diagnose UTIs; however, in smaller cities where there is no suitable facility for sample analysis, patients must rely entirely on their clinician's discretion. UTIs can be asymptomatic, acute, chronic, complicated, or uncomplicated, with clinical symptoms varying depending on the region of the urinary tract affected, the causative microorganisms, the severity of the disease, and the patient's ability to develop an immune response. 4 UTI symptoms include fever, burning sensation while urinating, lower abdominal pain, vaginal discomfort, genital, and supra-pubic pain.4 These symptoms vary depending on the person's age and the region of the infected urinary tract.5 Sex, age, ethnicity, circumcision,6, 7 HIV,8, 9, 10 diabetes, urinary catheterization, genitourinary tract anomalies,11, 12 sexual intercourse, pregnancy 13 and hospitalization status 14 are all factors that contribute to recurrent UTIs.15 Men have a lifetime risk of only 13%; however, women have a 60% probability of having UTI.16 Various parameters such as the type of UTI (complicated or uncomplicated), gender, age, and previous history of antibiotic medication of each UTI patient should be examined and considered to establish accurate global data on susceptibility.17

The increased accessibility of technological devices and user-friendly statistical software appears to have increased the usage of multivariable regression models in clinical research.18 Binary outcomes are common in biomedical research, and multivariable logistic regression is the most utilised regression model among the various binary link functions.19 A self-created questionnaire for the diagnosis of one of the most prevalent diseases, UTI, requires standardisation where validation is necessary to determine its accuracy with the help of microbiological culture. The study is quantitative in nature and focuses on the presence of UTI symptoms from the start. Its target population is a general population with basic traits who, when diagnosed with UTI, seek further speciality care if necessary. The objective of this study was to investigate the sensitivity and specificity of a newly designed questionnaire based on signs and symptoms through a personal interview. With this goal in mind, we address the following questions:

Materials and Methods

Results

Model fit assessment

According to the Hosmer-Lemeshow test (c2(8) = 2.156; p=0.976), the gap between observed and expected counts for the UTI is smaller (Table 4), and 82.2% of subjects were correctly classified (Table 5), implying that it is good model to fit the data. Also, the area under ROC (receiver operating characteristic) curve is 0.789 (95% CI: 0.74, 0.84) (Table 6), which is an acceptable fit; hence it is a good model to fit the data (Figure 1).

Table 4

Contingency for hosmer-lemeshow test

S.No.	UTI = No		UTI = Yes		Total
	Observed	Expected	Observed	Expected
1	22	22.000	4	4.000	26
2	21	22.063	29	27.937	50
3	10	9.949	24	24.051	34
4	14	14.404	45	44.596	59
5	10	8.364	31	32.636	41
6	7	6.864	48	48.136	55
7	3	4.266	40	38.734	43
8	5	3.490	40	41.510	45
9	2	2.098	48	47.902	50
10	0	.503	24	23.497	24

Table 5

Classification table of the preliminary final model

Observed count		Predicted
		UTI		Percentage Correct
		No	Yes
UTI	No	21	73	22.3
	Yes	3	330	99.1
Overall percentage				82.2

[i] The cut value is 0.500; the percentage correctly classified should be > 70% for a good model fit

Table 6

Area under the curve (AUC)

Area	Standard Errora	Asymptomatic significance b	Asymptomatic 95% CI
			Lower	Upper
0.789	0.026	0.000	0.737	0.840

The test result variable(s): Predicted probability has at least one tie between the positive actual state group and the negative actual state group where statistics may be biased; a- Under the nonparametric assumptions; b- Null hypothesis: true area = 0.5.

Figure 1

The ROC curve is the true positive rate (Sensitivity) is plotted in function of the false positive rate (100-Specificity) for different cut-off points

Discussion and Conclusion

The discovery of UTI-related parameters and their relationship to positive microbiological growth revealed a correlation between the onset of infection and microbiological culture reports. Our analysis demonstrated that the prevalence of UTI was 333/427 (77.9%). Our prevalence rate is higher than that of other researchers, which varies from 45.69%21 to 37.30%22 to 53.82%.17 The fundamental selection for this study is the presence of UTI symptoms in primary care; nevertheless, even though the data was limited and focused to the group with acute urinary symptoms, it can be used to build a UTI screening in the broader population. The high prevalence of UTIs in our study could be attributed to the rigorous initial screening as well as the inclusion of a range of risk categories such as diabetes, pregnancy, benign hypertrophy of the prostate, and others who are susceptible to UTIs. Even though diabetes, calculus, prostate enlargement, pregnancy, and other variables are known to predispose individuals to UTI, we found no evidence of this in our study since the number of patients with these predispositions and UTI was low. Our study correlates to a study conducted in Mexico 97.3%23 and Ethiopia 90.1%.24 A multinomial logistic regression25 analysis on UTI-related parameters was performed, with a 95% CI. During univariable analysis, ten independent variables were statistically significant, which were further put up for multivariable logistic regression analysis. During multivariable analysis, four independent variables (“urinary leakage because you cannot hold sudden desire to urinate”, “feeling of incomplete emptying of bladder”, “a desire to urinate when you have small amount of urine in bladder” and “pain in bladder or Lower abdomen”) were removed. We can conclude that six independent variables (“gender”, “how many times you urinate from morning till night”, “a sudden desire to urinate which is difficult to hold”, “weakness of urinary stream”, “splitting or spraying of urinary stream” and “fever”) were the final model. Gender has clinical implications for UTI since females are more susceptible to UTI than males, as demonstrated in this model. The proximity of the anus to the vagina and the shorter urethra may explain the greater prevalence of UTI in females. 4, 26, 27 The model has statistically demonstrated that the frequency of urination during the day increases from normal and that there is an urgent need to urinate, which becomes difficult to hold during UTI due to inflammation and irritation induced by bacteria in the bladder. However, depending on personal traits, bladder volume, and drinking habits, the frequency of regular urination might vary greatly from person to person. The frequency of micturition is commonly accompanied with dysuria due to UTI, but this was not the case in our study, thus we asked several other symptom-based questions about periodicity and others. A burning sensation or inflammation in the urethra may produce "weakness of the urinary stream" and "splitting or spraying of the urinary stream" during UTI; hence, to avoid pain, the external urethral sphincter voluntarily regulates urine flow from the bladder. In terms of model fit evaluation, the Hosmer-Lemeshow test, classification table, and area under the ROC curve show that our preliminary model has a good model fit. Each point on the ROC curve represents a sensitivity/specificity pair corresponding to a particular decision threshold. A test with perfect discrimination (no overlap in the two distributions) has a ROC curve that passes through the upper left corner (100% sensitivity and specificity). Therefore, the closer the ROC curve is to the upper left corner, the higher the test's overall accuracy 28. Therefore, to increase the rate of true positivity, one needs to have an even bigger sample size.

Urinary tract infections are a common infectious disease irrespective of gender and are becoming a major public health concern. A self-generated questionnaire for the diagnosis of UTI ensures standardization where specific pattern guideline is created that can be used as a marker and can be retested on a bigger population. The study focused on the urinary system as a whole and found that the recurrence of UTI is not always proportional to the severity of UTI due to a variety of factors such as immunity, individual experience and sensation, related symptoms, and so on. The logistic regression analysis showed a significant association between UTI and some of the variables. A statistical model (multiple logistic model) has also been established for the prediction of UTI and obtained its accuracy of prediction as 82.2% which can be instrumental for further treatment and prevention of UTI in this era of escalating Antimicrobial resistance (AMR).

Author Contribution Statement

Neelam Jain, Jagdish Prasad, Gopal Nath, Amit Bhargava and Kanika Bhargava hypothesised and designed the research plan. Amit Bhargav and Kanika Bhargava performed data acquisition. Jagdish Prasad, Alexandru-Atila Morlocan, Palak Khinvasara and Kanika Bhargava performed and analysed SPSS outcomes. Statistical analysis and preparation of the initial draft of the manuscript were done by Kanika Bhargava and Jagdish Prasad. Final editing and reviewing were done by Jagdish Prasad, Neelam Jain, Kanika Bhargava, Alexandru-Atila Morlocan, Gopal Nath, Amit Bhargava, Palak Khinvasara, Ragini Yadav and G. K. Aseri. All the authors have reviewed the manuscript.

Data Availability Statement

All data generated or analysed during the study are included in this manuscript.

Source of Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Conflict of Interest

The author(s) declare no competing interests

Acknowledgement

The authors are grateful to the hospital administration, Dean, OPD physicians, and technical staff of Hayes Memorial Mission Hospital S.H.U.A.T.S, Prayagraj (U.P.), India, for granting permission and assisting in the completion of the questionnaire for our study.