Ramadan and Kidney Disease (RaK) risk assessment tool. Potential risk calculator for evaluating the risk of Ramadan fasting in chronic kidney disease patients

Abstract

Background

Fasting during Ramadan has been found safe for most chronic kidney disease (CKD) patients, but may pose risks for some. The Ramadan and Kidney Disease (RaK) working group has developed a risk stratification tool to categorize CKD patients who intend to fast into low-, moderate-, and high-risk groups. We propose a Ramadan fasting calculator that incorporates key risk factors to guide fasting decisions for CKD patients.

Methods

A prospective cohort study was conducted at Ambulatory Healthcare Services from March 2024 to August 2024. The Electronic Medical Records (EMRs) of 220 patients were reviewed, supplemented by telephone interviews conducted before and after Ramadan. The primary outcomes assessed were the Ramadan fasting experience and the occurrence of adverse significant events (ASEs), which include hospital admissions.

Results

Among the 220 participants, 71.4% completed all 30 days of fasting during Ramadan, 7.3% began fasting but had to break it, and 21.4% did not fast. This sample’s mean and median RaK scores were 9.78 and 9.0, respectively; 55.5% had a score of 9 or lower, 24.5% had scores between 10 and 13, and 18.6% had scores above 13. A total of 39 adverse significant events (ASEs) were recorded. Using ordinal regression (proportional odds model) the RaK score was the only significant predictor of fasting (none, interrupted, complete), with an estimate of -0.159 (95% CI: -0.264 to -0.053), p = 0.003. The area under the curve (AUC) for predicting attempts at fasting (using logistic regression) was 0.715 (95% CI: 0.635–0.794), with an optimal cutoff of 9.5. Using logistic regression, a higher RaK score was significantly associated with an increased incidence of ASE (β = 0.253, OR = 1.287( 1.036–1.599, p = 0.023). Decision-curve analysis showed a positive net clinical benefit across threshold probabilities of 7–44%, identifying about 1 additional ASE patient per 10 evaluated patients and avoiding 1 unnecessary intervention per 7 patients compared with universal intervention at a 10% threshold.

Conclusion

The RaK score demonstrated potential predictive value for fasting status and adverse events among CKD patients; however, validation in different populations is recommended.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12882-026-05091-w.

Key learning points:

What was known:

Research on Pre-Ramadan risk assessment in relation to Ramadan fasting by kidney disease patients is lacking.

This study adds:

This is the first study to validate the RaK assessment tool developed based on multidisciplinary experts’ consensus.

The tool was effective in identifying patients at high risk of developing adverse events during fasting and in identifying patients who can fast safely.

Potential impact:

The RaK assessment tool is recommended to be included in the pre-Ramadan assessment of patients with kidney diseases.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12882-026-05091-w.

Introduction

Chronic Kidney Disease (CKD) doubled globally from 1990 to 2021, with significant geographical differences [1], and is now the seventh leading cause of mortality worldwide [2]. The contribution of CKD to mortality goes beyond that of End-Stage Kidney Disease (ESKD) [3], with recent projections suggesting that CKD-related deaths may double to 4 million by 2040 [2]. Implementing cost-effective interventions for CKD patients will reduce the risk of ESKD and cardiovascular diseases, thereby significantly alleviating the burden of CKD mortality.

Fasting during the holy month of Ramadan is one of the five pillars of Islam. Muslims must abstain from eating or drinking from dawn until sunset. Ramadan Fasting (RF) is reported to be safe for most kidney disease patients and may even contribute to moderate improvements in kidney function [4]; but this depends on CKD stage, comorbidities, and other factors, which may increase the risk of adverse health outcomes during fasting, particularly in patients with chronic kidney disease (CKD), who often have multiple comorbidities that complicate their management [5–7].

CKD is a significant global public health issue, affecting approximately 13% of the population worldwide [8]. Given its high prevalence and associated complications, CKD represents a substantial health burden both globally and regionally [9, 10]. In the United Arab Emirates, the estimated prevalence of CKD in Abu Dhabi is around 8%. This prevalence is higher than that reported elsewhere in the Gulf region- for example, in a cross-sectional study in Saudi Arabia, CKD prevalence was found to be 4.76% -but differences in study design could influence differences in reported prevalence [11].

Muslims with illness are exempted from fasting, but decision-making by physicians and patients is complex, as the risk of harm cannot be predicted with certainty. To provide individualized recommendations for patients with CKD, accurate risk assessment is essential. While a risk stratification tool for identifying high-risk diabetes patients has been developed and validated by the IDF-DAR, no such tool previously existed for CKD patients [12] The Ramadan and Kidney Disease Initiative (RaK) was established in May 2022 to assist healthcare teams in decision-making regarding fasting of CKD patients,. This initiative brought together an international panel of experts, including nephrologists, endocrinologists, and family medicine physicians with expertise in CKD and Ramadan fasting. Boobes et al. [4] developed a risk assessment and stratification tool, along with clinical recommendations, to guide fasting decisions for CKD patients and their healthcare providers [4, 13].

To develop these guidelines, the RaK group conducted a systematic review of studies on Ramadan fasting and its associated risks. This review led to the development of a scoring system (tool) that stratifies patients into risk categories, aiding counseling on fasting during Ramadan. The RaK tool included age, GFR level, level of Proteinuria, presence of diabetes, glycemic control, hypertension, presence of CVD, history of a kidney transplant, history of kidney stones, medications used (CNI, Diuretics, SGLT2 I, ARB/ACE-I), country duration of fasting in hours, patient exposure to temperature, physical labor, past fasting experience, and history of acute kidney injury during the last 3 months.

Due to the limited availability of quantitative studies in this area, most of the scoring system’s development was based on expert consensus. This study is the first to validate the RaK scoring tool for risk stratification and shared decision-making among CKD patients considering Ramadan fasting.

Method

Study design

This prospective cohort study was conducted at Abu Dhabi SEHA Clinics (SC) centers from March to August 2024.

Study population

The source of patients was the SC CKD Population Health Program, which was implemented in 2018 at SC centers. An EMR report is generated monthly for all lab results of SC patients with abnormal kidney function tests, which are reported to SC centers for follow-up and management.

Inclusion and exclusion criteria

The exposure window for all participants was the single month of Ramadan (March 2024). The study included CKD patients aged 18 years or older attending SEHA Clinics (SC) centers who were either UAE nationals or foreign residents of the Abu Dhabi Emirate, with a diagnosis of CKD according to the Kidney Disease Improving Global Outcomes (KDIGO) CKD guidelines [14]. Exclusion criteria were patients with acute kidney injury not related to CKD, inability to provide consent, and non-Arabic/English speakers without translation.

As per supplementary 1, patients who appeared in the report between October and March 2024 were included in this study, resulting in 450 eligible patients. Due to time and other resource constraints, only 330 were contacted, of whom 271 were reachable and agreed to participate. Only 220 accurately reported their fasting status, and those who fasted or attempted fasting were included in the analysis. The 51 missing fasting statuses had given consent and participated in answering the questionnaire before Ramadan, but were unreachable or did not answer the calls (three separate days of calls) after Ramadan. There was no significant difference in age, with an average age of 66 years among participants compared to 68.1 years among non-participants, sex, 66.6% females among participants compared to 54.1% among the non-participants, or race, 60% UAE nationals among participants compared to 66.6% UAE nationals in the non-participants (p-values 0.27, 0.56, and 0.56, respectively). While participants may well have differed from non-participants in unmeasured characteristics (selection bias), it seems unlikely that included and excluded patients had meaningfully different risk profiles, as participants were not selected based on clinical severity or any outcome-related characteristics.

Among the 220 included in the final analysis, 173 fasted at least one day, while 47 did not attempt to fast at all, supplementary 1.

Data collection

Participants were called and interviewed by telephone both before and after Ramadan. Consent to participate was sought from each participant, and an EMR audit was conducted for all those agreeing to participate after each interview. The first tele-interview was in March 2024, before Ramadan (March 10-April 9, 2024). CKD patients were contacted and assessed using the RaK proposed risk assessment tool (supplementary 2). Frailty assessment was performed for patients aged 60 and older using the FRAIL tool, a validated questionnaire with five questions that predict clinical outcomes [15, 16]. These five questions were related to fatigue, resistance, or climbing stairs, ambulation (i.e., walking a couple of blocks, number of chronic illnesses, and losing weight by more than 5%. Data collection was conducted by family medicine residents who were trained to deliver these interviews as teleconsultations as part of their training. The interviews lasted less than 10 min each.

Outcome assessment

There were two primary outcomes: the Ramadan fasting experience and the occurrence of adverse significant events (ASEs). The Ramadan fasting experience was categorized as an ordinal outcome: patients who did not fast any day, patients who attempted to fast but had to break their fast (fasted at least a day), and patients who completed the fast.

The outcome was assessed after Ramadan, using additional data to identify significant outcomes, such as illnesses, admissions, or health issues that developed during fasting. These include admissions recorded in the EMR or self-reported admissions, and symptoms severe enough to break fast or require reporting, such as hypoglycemia, fever, dizziness, or chest pain. Only events occurring during Ramadan were considered. The surveillance started on the first day of Ramadan and continued until the end of the holy month. Any deterioration in renal function was assessed within 3 months after Ramadan.

Data analysis

Sample size

To assess the statistical power of our study, we considered a scenario in which 25% of the patients would have a specific event, cases, and 50% of the, controls, would be exposed to a binary risk factor. Then, a sample of 220 would be able to demonstrate an odds ratio of 2.5 for the association between that risk factor and the event with 80% power and 5% significance level.

Statistical analysis was done with SPSS v29. In view of the relatively small sample size, which may reduce the reliability of asymptotic methods, we assessed parameter-estimate biases and stability in all analyses using 5000 bootstrap samples with bias-corrected and accelerated (BCa) 95% confidence intervals. Fasting status was treated as an ordinal dependent variable. Ordinary regression analysis was used to assess if the baseline variables predicted fasting status; not fasting, attempting to fast but stopped fasting, or fasted all Ramadan days. ASE predictions were analyzed using logistic regression. To assess potential measurement bias arising from the composite endpoint, sensitivity analyses were performed using (1) hospital admissions alone and (2) the composite endpoint, including hospital admissions and self-reported adverse events. Other variables were treated as independent variables, including the RaK risk category. Frequencies, cross-tabulation, proportional odds, logistic, and linear regression were used. C-statistics were used to assess the performance of the prediction model derived from the regression analysis. Decision-curve analysis was performed using the R package dcurves to estimate the net benefit of the RaK Score across threshold probabilities.

There was no exclusion of subjects based on their fasting choices in Ramadan. This choice was a future decision, and non-fasting participants represent a risk category informative for the RaK score’s performance. As well, household Ramadan routines, disrupted sleep-wake cycles, altered meal timing, and reduced daytime activity also produce relevant exposures regardless of fasting. Additional sensitivity analysis, including fasting behavior, was done to assess if the ASE binary regression changes with the addition of fasting status.

The thresholds (< 5, < 9, 9–13, > 13) followed the original score’s guidelines. The Youden index was additionally computed across candidate cutoffs to describe RaK score performance in the study. We performed 5000 stratified bootstrap resamples in R version 4.5.1, GUI 1.82, using the pROC package to estimate the 95th-percentile confidence interval for the Youden-optimal cutoff.

Ethical approval and consent to participate

The study was approved by the Ambulatory Healthcare Services IRB, approval number 2024-5, dated March 12^th, 2024. All methods were carried out under relevant guidelines and regulations. The authors confirm that the study was conducted in accordance with the Helsinki Declaration.

Patient and public involvement

Patients or the public WERE NOT involved in the design, or conduct, or reporting, or dissemination plans of our research.

Results

Participants’ fasting experience and description

Fasting Out of 220 individuals, 71.4% completed all 30 days of fasting during Ramadan, while 7.3% started or attempted to fast but had to break their fast, and 21.4% did not fast at all. There were 39 significant adverse events: 12 (30.8%) among those who did not fast, 8 (20.5%) among those attempting to fast, and 19 (48.7%) among those who completed their fast. All 39 ASEs are listed in Table 1, with 30 requiring admissions. The 9 without admission were sufficiently severe to break the fast: hypoglycemia, fever, and chest pain.

Table 1.

A. The relation between Fasting Ramadan status and the occurrence of Adverse Significant Event ASE

	Did not fast single day	Fasted some days	Fasted all days	Total	Chi-square	P value
No events	35	8	134	177	16.2	< 0.001
	74.50%	50.00%	87.90%	82.70%
Significant events	12	8	17	37
	25.50%	50.00%	12.10%	17.30%
	0	1	2

B. Reported Adverse Significant Event ASE distributed by RAK Score

	RAK Score categories			Total
	<=9	10–13	> 13
No Adverse Significant Events (ASEs)*	109	40	28	179
	90.8%	75.5%	68.3%	82.7%
Adverse Significant Events (ASEs)*	11	13	13	38
	9.2%	24.5%	31.7%	17.3%
	122	54	41	214
	56.2%	24.9%	18.9%	100.0%

*Reported diagnosis of Adverse Significant Events

Significant events with no admission	Count
Fatigue	1
Flank pain with elevated creatinine levels	2
Chest pain	1
Fever	5
Significant events with admission
Gangrene of the right little toe requiring toe amputation at the metatarsophalangeal joint	1
Pneumonia	2
Acute kidney injury (AKI)	4
Acute kidney injury with congestive heart failure
Aneurysm embolization	1
Congestive heart failure	1
Chest pain	1
Fracture	2
Day-case left lumpectomy and sentinel lymph node biopsy	2
Dialysis	2
Drainage of right upper thigh abscess with excision of hidradenitis	1
Dyspnea	2
Fatigue	3
Fever	1
Hyponatremia	1
Nausea and vomiting	1
Neck pain and dizziness	1
Acute coronary syndrome	1
Stroke	1
Urinary obstruction secondary to benign prostatic hyperplasia (BPH)	1
Urinary tract infection (UTI) with sepsis	1
	30

Our sample included many high-risk patients: 67.3%were 60 or older, nine patients had a history of transplant, four were on dialysis, 20.1% on insulin, 56.8% were diabetics, 7.3% experienced AKI in the last 3 months before Ramadan, 25% were frail, and 13.6% had a hospital admission within 3 months before Ramadan. Table 2 presents more characteristics of the sample.

Table 2.

Subjects characteristics distributed by the fasting status

		Did not fast single day	Fasted some days	Fasted all days	Total	P V alue
Age groups	<=30	0	0	4	4	0.239
		0.00%	0.00%	2.50%	1.80%
	31–40	6	2	8	16
		12.80%	12.50%	5.10%	7.30%
	41–50	3	0	19	22
		6.40%	0.00%	12.10%	10.00%
	51–60	5	1	24	30
		10.60%	6.30%	15.30%	13.60%
	61–70	7	4	41	52
		14.90%	25.00%	26.10%	23.60%
	70–80	15	6	41	62
		31.90%	37.50%	26.10%	28.20%
	> 80	11	3	20	34
		23.40%	18.80%	12.70%	15.50%
Sex		0	1	2		0.49
	Female	25	6	83	114
		53.20%	37.50%	52.90%	51.80%
	Male	22	10	74	106
		46.80%	62.50%	47.10%	48.20%
UAE citizenship	0	20	10	59	89	0.15
	Non UAE	20	10	59	89
		42.60%	62.50%	37.60%	40.50%
	UAE	27	6	98	131
		57.40%	37.50%	62.40%	59.50%
Dialysis		0	1	2		0.43
	No	44	15	137	196
		97.80%	93.80%	98.60%	98.00%
	Yes	1	1	2	4
		2.20%	6.30%	1.40%	2.00%
Diabets Melitus		0	1	2		0.27
	No diabetes mellitus	16	6	70	92
		34.00%	37.50%	44.60%	41.80%
	Type 1 Diabets Melitus	2	0	1	3
		4.30%	0.00%	0.60%	1.40%
	Type 2 Diabets Melitus	29	10	86	125
		61.70%	62.50%	54.80%	56.80%
Cardiovasular Diseases		0	1	2		0.094
	No heart diseases	24	7	107	138
		51.10%	43.80%	69.50%	63.60%
	Stable Multi-Vessel Disease	17	8	40	65
		36.20%	50.00%	26.00%	30.00%
	Unstable Multi-Vessel Disease	5	1	6	12
		10.60%	6.30%	3.90%	5.50%
	Compensated Congestive heart failure	0	0	1	1
		0.00%	0.00%	0.60%	0.50%
	Cardiac arrhythmia	1	0	0	1
		2.10%	0.00%	0.00%	0.50%
On Diuretics		0	1	2		0.025
	No	27	9	120	156
		58.70%	56.30%	76.40%	71.20%
	Yes	19	7	37	63
		41.30%	43.80%	23.60%	28.80%
On SGLT2 i		0	1	2		0.97
	No	30	11	103	144
		65.20%	68.80%	65.60%	65.80%
	Yes	16	5	54	75
		34.80%	31.30%	34.40%	34.20%
On Insulin		0	1	2		0.004
	No	29	15	131	175
		63.00%	93.80%	83.40%	79.90%
	Yes	17	1	26	44
		37.00%	6.30%	16.60%	20.10%
On Glucocorticoids		0	1	2		0.002
	No	33	4	107	144
		71.70%	25.00%	68.20%	65.80%
	Yes	13	12	50	75
		28.30%	75.00%	31.80%	34.20%
On Sulfonylureas		0	1	2		0.029
	No	44	15	127	186
		95.70%	93.80%	80.90%	84.90%
	Yes	2	1	30	33
		4.30%	6.30%	19.10%	15.10%
On Angiotensin-converting enzyme inhibitors		0	1	2		0.011
	No	33	16	100	149
		15.10%	7.30%	45.90%	68.30%
	Yes	13	0	56	69
		28.30%	0.00%	35.90%	31.70%
Labor		0	1	2		0.29
	0	41	14	125	180
		18.60%	6.40%	56.80%	81.80%
	1	5	1	30	36
		2.30%	0.50%	13.60%	16.40%
	2	1	1	2	4
		0.50%	0.50%	0.90%	1.80%
History of Kidney Transplant		0	1	2		0.085
	0	44	14	153	211
		20.40%	6.50%	70.80%	97.70%
	0.5	1	0	0	1
		0.50%	0.00%	0.00%	0.50%
	1	1	0	0	1
		0.50%	0.00%	0.00%	0.50%
	2	0	1	2	3
		0.00%	0.50%	0.90%	1.40%
Kidney transplant		0	1	2		0.107
	No	44	14	153	211
		20.00%	6.40%	69.50%	95.90%
	Yes	3	2	4	9
		6.40%	12.50%	2.50%	4.10%
HO Kidney stones		0	1	2		0.946
	No	45	15	149	209
		20.50%	6.80%	67.70%	95.00%
	Yes	2	1	8	11
		0.90%	0.50%	3.60%	5.00%
Twice daily Calcineurin inhibitors		0	1	2		0.277
	No	43	14	151	208
		19.60%	6.40%	68.90%	95.00%
	Yes	3	2	6	11
		1.40%	0.90%	2.70%	5.00%
Hospital admission within 3 months		0	1	2		0.13
	No	36	13	135	184
		16.90%	6.10%	63.40%	86.40%
	Yes	10	3	16	29
		4.70%	1.40%	7.50%	13.60%
Acute kidney injury during last 3 months		0	1	2		0.034
	No	28	9	115	152
		17.10%	5.50%	70.10%	92.70%
	Yes	3	3	6	12
		1.80%	1.80%	3.70%	7.30%
Frailty status		0	1	2		0.023
	Robust	22	9	75	106
		10.00%	4.10%	34.10%	48.20%
	Pre-Frail	6	3	50	59
		2.70%	1.40%	22.70%	26.80%
	Frail	19	4	32	55
		40.40%	25.00%	20.40%	25.00%
DYSLIPIDEMIA		0	1	2		0.93
	No	20	6	63	89
		9.10%	2.70%	28.60%	40.50%
	Yes	27	10	94	131
		12.30%	4.50%	42.70%	59.50%
Hypertension		0	1	2		0.14
	No	1	2	18	21
		0.50%	0.90%	8.30%	9.70%
	Yes	46	14	135	195
		21.30%	6.50%	62.50%	90.30%
Total		47	16	157	220

Validating the RaK tool

The sample had a mean RaK score of 9.78 and a median of 9; 55.5% had a RaK score of 9 or less, 24.5% 10–13, and 18.6% more than 13. Figure 1 shows the distribution of RaK scores in the study sample by Ramadan fasting status and ASE status (mean of 12 points among patients with a score > 13, 10 among patients with a score of 10–13 points, and 8 among subjects with a RaK score < 10 points). Patients who did not fast a single day (n = 47) had higher mean RaK scores (12.49 vs. 8.79) and lower mean eGFR (33.64 vs. 45.19) than those who tried (n = 173); the score thus correctly identified higher-risk patients who ultimately did not fast.

Fig. 1.

RaK score distribution by fasting status and occurrence of significant adverse events

The two primary outcomes were analyzed using regression analysis with respect to the RaK tool and fasting status. Ordinal regression was used to explore predictors of fasting status, with groups defined as: did not fast any day (y = 0), fasted some days (y = 1), or fasted all days (y = 3) (Table 3). The independent variables considered were all the variables collected in the study.

Table 3.

Prediction of the ability to fast using ordinal regression, with attempting fasting all days as the reference, compared to attempting some days or none of the days

		Estimate	P value	95% Confidence Interval
Threshold	[Did not attempte to fast any day( 0] *	-2.705	0.023	-5.032	-0.379
	Attempted fasting some days (1) *	-2.297	0.052	-4.615	0.021
Location	RAK Score	-0.159	0.003	-0.264	-0.053
	Age (years)	-0.006	0.583	-0.028	0.016
	eGFR	0.013	0.364	-0.015	0.041
	Female = 0]	0.255	0.441	-0.393	0.903
	[Male = 1]	0a	.	.	.

*Reference is fasting all days of Ramadan (2)

Predicting Ramadan fasting

Only the RaK risk score was a significant predictor, with an estimate of -0.159 (95% CI: -0.264 to -0.053), p-value = 0.003. That is, for each one-unit increase in the RaK score, the odds of fasting during Ramadan decrease by 15%. Bootstrapped BCa 95% CI confirmed stability of the 95% CI estimate (-0.275 to -0.054). Bootstrap bias estimates were minimal across all parameters (range − 0.065 to 0.007). The AUC for predicting attempting fasting was 0.715 (0.635–0.794), with the best cutoff point of 9.5, a sensitivity of 69.8%, and a specificity of 66.9%. Figure 2. A sensitivity analysis was performed, excluding patients with kidney transplants; the results did not change.

Fig. 2.

ROC curves for the RaK Score (A) Predicting Adverse Significant Events (ASE) during Ramadan (B) Predicting fasting behavior: any attempt to fast (blue), completed full month (red)

Predicting adverse significant events

In the multivariable logistic regression analysis, two baseline characteristics were independently associated with ASE, higher RaK Score and friality. A higher RaK Score was associated with greater odds of ASE, with each 1-point increase corresponding to a 12.9% increase in the odds of ASE, 1.129 (1.039–1.227), as shown in Table 4; Fig. 3. Although frailty status was a significant predictor overall, its effect was concentrated at the frail category. Being frail compared to robust was associated with nearly three-fold higher odds of ASE (OR 2.748, 95% CI 1.165–6.483, p = 0.021), whereas being pre-frail compared to robust showed no significant increase in risk (OR 0.991, 95% CI 0.359–2.731, p = 0.986). Worth noting that although nationality (UAE vs. non-UAE) was identified as a significant predictor of ASE, it was not included in the final model. This is because it represents a more context-specific, rather than generalizable, biological risk and therefore may not be transferable to other healthcare settings (Table 4).

Table 4.

Multivariable logistic regression for predicting acute symptomatic events (ASE) during Ramadan

Variable	B	Wald	p-value	OR (95% CI)
A. Primary multivariable logistic regression model
RAK Score (per one-point increase)	0.122	8.21	0.004	1.129 (1.039–1.227)
FRAIL Score (overall, 2 df)	—	5.79	0.055	—
Pre-frail vs. robust (reference)	−0.009	0.00	0.986	0.99 (0.359–2.731)
Frail vs. robust (reference)	1.011	5.30	0.021	2.748 (1.165–6.483)
B. Sensitivity analysis with fasting behaviour added
RAK Score (per one-point increase)	0.099	4.49	0.034	1.10 (1.008–1.211)
FRAIL Score (overall, 2 df)	—	6.39	0.041	—
Pre-frail vs. robust (reference)	0.112	0.04	0.839	1.12 (0.380–3.290)
Frail vs. robust (reference)	1.130	5.92	0.015	3.10 (1.251–7.653)
Fasting behaviour (overall, 2 df)	—	8.65	0.013	—
Attempted but did not complete vs. did not fast (ref.)	1.563	5.40	0.020	4.78 (1.282–17.785)
Completed full month vs. did not fast (reference)	−0.421	0.75	0.386	0.66 (0.254–1.699)

A: presents the primary multivariable model used to validate the RaK Score; only baseline predictors ascertained before Ramadan are included. B: is a pre-specified sensitivity analysis adding fasting behavior as a three-level categorical covariate (did not fast [reference]; attempted but did not complete; completed the full month) to test whether the RaK Score’s predictive value is independent of fasting behavior

FRAIL Score entered as a 3-level categorical variable (0 = robust [reference], 1 = pre-frail, 2 = frail);

* The RAK Score × fasting behavior interaction was not statistically significant (p = 0.54), indicating that the score’s per-point effect is consistent across all three pre-Ramadan decision groups

Fig. 3.

Predicted probability of adverse events in relation to RaK score

Two statistical tests were performed to examine the derived ASE determinants model. First, multicollinearity among the predictors was evaluated by obtaining the Variance Inflation Factor (VIF) and Tolerance statistics, which were: RAK Score Tolerance = 0.908, VIF = 1.10, pre-frail (Tolerance = 0.872, VIF = 1.15, and frail (Tolerance = 0.800, VIF = 1.25. Confirming the absence of multicollinearity.

Second, 95% bootstrap confidence intervals (Bca) were computed using 5000 resamples that showed minimal bias (range − 0.051 to + 0.031), and the BCa 95% confidence intervals for the regression coefficients excluded zero for both significant predictors: RaK Score (B = 0.122, BCa 95% CI 0.031 to 0.206) and FRAIL frail vs. robust (B = 1.011, BCa 95% CI 0.060 to 1.924). The BCa interval for the pre-frail vs. robust contrast crossed zero (B = − 0.009, BCa 95% CI − 1.111 to 1.110), supporting the regression finding that pre-frailty does not add independent prognostic information beyond robust status in this cohort.

Discrimination and threshold analysis

The RaK Score discrimination between participants with and without ASE was acceptable. The AUC was 0.701 (95% CI 0.614–0.789) with a cutoff of 8.5 identified by the Youden index. The sensitivity 83.8% and the specificity 50.8%; Supplement 3a. A 5000 stratified bootstrap resample showed the median bootstrap-derived optimal cutoff was 8.5, with a 95% bootstrap percentile interval of 7.5 to 11.5. This interval includes the pre-specified published RaK Score threshold of 9, supporting the published threshold. By comparison, eGFR alone performed poorly (AUC 0.590, 95% CI 0.489–0.691), confirming that the composite RaK Score provides superior risk stratification compared with renal function alone.

The pre-Ramadan RaK Score was higher among participants who experienced an ASE than among those who did not (Fig. 3, and this difference was also related to fasting status. Among participants who experienced an ASE during Ramadan, 7 of 8 (87.5%) who attempted but did not complete the full month of fasting had a RaK Score in the moderate-to-high range ≥ 9, and 44.1% of those who fasted the full month had a moderate-to-high score ≥ 9, including 16.7% at the high-risk threshold > 13. In contrast, among participants without an ASE who completed the full month of fasting, 31.6% had a moderate-to-high score ≥ 9, and only 9.6% crossed the high-risk threshold. These descriptive distributions are consistent with the multivariable findings: the risk of ASE increased with increasing RaK Score, rising from approximately 6% at low score levels to over 50% at high score levels, Fig. 4.

Fig. 4.

RaK score categories among patients who had significant adverse events and those who did not distributed by fasting status

Calibration of the RAK Score

The RaK Score model was well calibrated. The overall observed-to-expected event ratio was 1.00 (37 observed versus 37.0 expected ASE events across the cohort), and the Hosmer–Lemeshow test was not significant (χ² = 5.10, df = 8, p = 0.747). Additionally, decision-curve analysis, DCA, Fig. 5, Supplement 3, demonstrated that the RaK Score yielded positive net clinical benefit relative to both ‘treat all’ and ‘treat none’ strategies across threshold probabilities of approximately 8% to 22%, which is the range most needed pre-Ramadan for clinical decision-making due to uncertainity in the low-to-moderate risk patients rather than for high-risk patients whose status is probably clinically evident.

Fig. 5.

Decision-curve analysis of the RaK Score showing net clinical benefit in relation to both “treat all” and “treat none” strategies across threshold probabilities of 8–22%

To test the possible effect of fasting exposure, a sensitivity analysis was conducted by adding fasting behavior (did not fast/attempted but did not complete/completed a full month) as a categorical covariate. Patients who did not fast had no difference from those fasted the whole month with regards to the ocuranc eof ASE, while those who attempted fasting had signfiicantly more risk of ASE compared to those who did not fast. The RaAK Score and frailty (frail vs. robust OR 3.10, p = 0.015) were significant in the sensitivty analysis model, which is similar to the primary model. The interaction between RaK Score and fasting behavior was not statistically significant (p = 0.54), indicating that RaK score performance is consistent across all three pre-Ramadan decision groups, Supplement 3.

Discussion

This cohort included many high-risk patients in an outpatient setting, including those with cardiac diseases, diabetes, hypertension, Proteinuria, and other comorbidities. This diversity strengthens both the internal and external validity of our study, which is the first validation study of the RaK score. It demonstrates the RaK Score’s ability to predict ASE across the full clinical decision spectrum, including patients who decided against fasting, those who attempted and stopped, and those who completed the full month of fasting. Additionally, the RaK score correctly identified the higher-risk patients who ultimately did not fast. This supports the RaK score as a clinically meaningful risk stratification tool rather than a proxy for physician behavior alone. Nevertheless, further research is needed to assess its applicability in special risk groups and substantially different populations. For instance, the growing number of dialysis patients and kidney transplant recipients who wish to fast represents a crucial area for further investigation.

Currently, the IDF-DAR tool is the only validated risk assessment tool for Ramadan fasting. Since nearly half of the participants in our study had diabetes, the RaK tool could be used in conjunction with the IDF-DAR tool for CKD patients with diabetes who plan to fast.

The ability to fast is closely tied to a patient’s overall health and well-being, making it a valuable outcome measure for validation. Statistical evidence suggests that a RaK score of ≤ 9 indicates low risk, 10–13 indicates moderate risk, and > 13 indicates high risk for ASEs. Specifically, this categorization was based on the observation that only 9.6% of patients with a RaK score > 13 fasted throughout Ramadan without experiencing any ASEs, while 68.4% of patients who completed fasting without ASEs had a RaK score of 9 or lower. Beyond quantifiable risk, an essential factor in decision-making is the individual’s acceptable level of harm as perceived by both healthcare providers and patients, and may be affected by factors related to patients’ social determinants of health and healthcare system resources. The occurrence of ASEs was significantly associated with frailty status. Similar studies [17–19] have also highlighted the role of frailty, emphasizing the need to strengthen the healthcare system through interventions and care guidelines focused on early detection and management. Other assessments of older patients can be assessed in future studies, such as Basic (BADL) and Instrumental (IADL) Activities of Daily Living scores [20]. With regard to the RaK tool specifically, future revisions of the score may benefit from assigning greater weight to the frail category, while the pre-frail category may not require additional risk attribution.

Although eGFR performance was acceptable in predicting fasting status, it was less effective in predicting ASEs, which may be due to a bias in which physicians advise patients with lower eGFR not to fast. However, confirming this hypothesis is challenging because randomization would be unethical, and patient preferences and adherence to medical advice are difficult to control. Additionally, for fasting, the model performs acceptably in supporting clinical decision-making. Nevertheless, for predicting ASEs, the model’s moderate specificity (50.9%) may result in substantial false-positive and false-negative rates. While the RaK model performed far better than using eGFR alone, the clinical consequence of a high false-positive rate may potentially discourage many low-risk patients from fasting. As with many similar risk assessment tools, this tool is not to be used as a stand-alone decision aid but as a screening tool to inform clinical judgment in the context of the specific patient’s condition. Proper counseling skills are needed for shared decision-making. In our cohort, the AUC of the RaK Score (0.701, 95% CI 0.614–0.789) was higher than that of the IDF-DAR tool (0.610) for predicting ASEs, suggesting that the integration of frailty and broader clinical context contributes meaningful incremental value over the IDF-DAR’s narrower variable set [21]. Future studies can further improve specificity by incorporating memberships of special populations and social determinants of health. Finally, the cutoff identified in this study (8.5) was close to the published threshold of 9, and the data’s bootstrap interval (7.5 to 11.5) included this threshold; we do not advise changing the RaK Score thresholds based on a single-center validation. Threshold optimization should be informed by larger multicentre cohorts with formal recalibration. However, this validation is the most comprehensive to date, offering valuable evidence for the ongoing RaK initiative as more data from other centers become available.

Interestingly, the Decision-curve analysis demonstrated that the RaK Score showed a positive net clinical benefit across a clinically relevant range of approximately 7% to 44%, the range that corresponds to realistic pre-Ramadan decision-making for intensive counseling, medication adjustment, and structured monitoring. For every 10 patients evaluated with the RaK Score, the score identifies approximately one additional patient with ASE beyond what would be detected by intervening on no one, after accounting for the harm of false positives at this threshold. Compared with a policy of intervening on all patients with chronic kidney disease attending pre-Ramadan assessment, the RaK Score allows approximately one unnecessary intervention to be avoided for every 7 patients evaluated, preserving the autonomy and religious observance of patients at lower risk while still flagging those who would benefit from individualized counseling.

Additionally, implementing this tool has implications that extend beyond improving care outcomes; it also supports physicians’ decision-making, enhances their knowledge, and promotes greater patient engagement in their care. As risk factor distributions and treatment patterns change, predictive performance may change over time. Therefore, regular recalibration and model updates using current cohorts will be necessary to maintain accuracy and reliability. Prospective validation and potential integration into digital clinical decision-support systems could improve scalability and clinical usefulness.

Limitations and potential biases

Although this study’s strengths include its prospective design and community-based sample, it relies on self-reported fasting status and adverse events, is limited to a single geographic population (Abu Dhabi), and includes small subgroups (transplant and dialysis). These factors affect generalizability and recall reliability. Additionally, using the two primary outcomes, fasting ability and ASEs, both have limitations. There is potential discordance between the ability to fast and medical safety during fasting. While fasting completion indicates patient tolerance and adherence to behavior, it does not necessarily reflect physiological stability. Some individuals may finish Ramadan despite experiencing clinically significant metabolic fluctuations, while others may stop fasting due to mild or precautionary symptoms. Therefore, fasting experience should be seen as a patient-centered measure of tolerance rather than a direct indicator of medical safety.

Regarding ASEs, although most were hospital admissions, a few were self-reported. Perceptions of symptoms and thresholds for interrupting fasting may vary among individuals and cultures. Furthermore, this is an observational study; thus, residual confounding cannot be ruled out despite multivariable adjustments. Another limitation is that the secondary risk-factor finding for frailty is best interpreted as an observation from the primary analytic cohort and warrants confirmation in larger studies. Socio-economic factors need to be included in future studies to improve score performance, as nationality showed a significant association, which may indicate contextual and setting factors that can not be ignored.

A further limitation concerns the partial self-fulfilling nature of risk-score validation in clinical practice. The RaK Score includes components (such as low eGFR, kidney transplant status, and active cardiovascular disease) that also influence physician advice regarding the safety of fasting. To the extent that high-risk patients receive and follow physicians’ advice not to fast, the score’s apparent predictive value partly reflects clinical decision-making rather than purely biological risk. We addressed this by validating the score across the full pre-Ramadan decision spectrum, comprising patients who decided against fasting, attempted but did not complete fasting, and completed the full month, rather than restricting to a single decision subgroup. A formal sensitivity analysis that added fasting behavior to the model confirmed that the RaK Score’s effect on ASE odds is consistent across all three decision groups. However, documented physician advice and intermediate clinical actions (e.g., dose adjustments, monitoring intensity) were not consistently recorded in the medical record for all participants and could not be entered as covariates. Future prospective validation studies should capture physician advice and clinical actions as explicit variables to allow direct adjustment for these mediators of the score-outcome relationship.

Conclusion

The RaK score demonstrated potential predictive value for fasting status and adverse events among CKD patients. The RaK score should be understood as a tool to inform and structure physician-patient counseling for individualized fasting guidance. Yet, external validation in different populations is recommended.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

LBK and YB conceptualized the study. LBK analyzed the data, and NN reviewed the analysis. LBK wrote the manuscript. EM, FM, MY, JS, AK, AA, AB, AS, SS, RK, MA, FK, AA, TF, WA, NY, HA collected data. All authors reviewed and approved the final manuscript.

Funding

No funding was used in this study.

Data availability

Data access is restricted as it is subject to the institution’s data access policy.

Declarations

Ethical approval and consent to participate

The study was approved by the Ambulatory Healthcare Services IRB, approval number 2024-5, dated March 12^th, 2024. All methods were carried out under relevant guidelines and regulations. The authors confirm that the study was conducted in accordance with the Helsinki Declaration.

Informed consent

Informed consent was gathered verbally during the two occasions of calling patients before and after Ramadan.

Consent to publish

Not applicable.

Competing interests

The authors declare no competing interests.