This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR mHealth and uHealth, is properly cited. The complete bibliographic information, a link to the original publication on https://mhealth.jmir.org/, as well as this copyright and license information must be included.
Chronic kidney disease (CKD) is a global health burden. However, the efficacy of different modes of eHealth care in facilitating self-management for patients with CKD is unclear.
The aim of this study was to evaluate the effectiveness of a mobile app–based intelligent care system in improving the kidney outcomes of patients with CKD.
Our study was a retrospective analysis based on the KidneyOnline intelligent system developed in China. Patients with CKD but not dependent on dialysis who registered on the KidneyOnline app between January 2017 and January 2021 were screened. Patients in the the KidneyOnline intelligent system group and those in the conventional care group were 1:1 matched according to their baseline characteristics. The intervention group received center-based follow-up combined with the KidneyOnline intelligent patient care system, which was a nurse-led, patient-oriented collaborative management system. Health-related data uploaded by the patients were integrated using deep learning optical character recognition (OCR). Artificial intelligence (AI)–generated personalized recipes, lifestyle intervention suggestions, early warnings, real-time questions and answers, and personalized follow-up plans were also provided. Patients in the conventional group could get professional suggestions from the nephrologists through regular clinical visits, but they did not have access to the service provided by AI and the health coach team. Patients were followed for at least 3 months after recruitment or until death or start of renal replacement therapy.
A total of 2060 eligible patients who registered on the KidneyOnline app from 2017 to 2021 were enrolled for the analysis. Of those, 902 (43.8%) patients were assessed for survival analysis after propensity score matching, with 451(50%) patients in the KidneyOnline intelligent patient care system group and 451(50%) patients in the conventional care group. After a mean follow-up period of 15.8 (SD 9.5) months, the primary composite kidney outcome occurred in 28 (6%) participants in the KidneyOnline intelligent patient care system group and 32 (7%) in the conventional care group, with a hazard ratio of 0.391 (95% CI 0.231-0.660;
The utilization of the KidneyOnline intelligent care system was associated with reduced risk of unfavorable kidney outcomes in nondiabetic patients with CKD.
Chronic kidney disease (CKD) is a worldwide public health concern and is increasingly becoming a global economic burden [
There has been growing interest from physicians and other health care providers to use novel health-related mobile apps for patients in their fields of interest. Telehealth educational apps are more flexible and adaptable to patients' preferences than paper-based or in-person health education. However, according to a systematic review, CKD-related apps accounted for only 1% of the total number of available apps in 2017 [
Up to now, most studies describing the effectiveness of eHealth interventions on patients with CKD were from North America [
This study was a retrospective cohort analysis based on a mobile app called KidneyOnline in China. The KidneyOnline app offered a platform with free materials and resources about CKD. Patients received recruitment information about the KidneyOnline app within WeChat; subsequently, they downloaded the app and became active app users. The KidneyOnline intelligent patient care system was embedded in the app, and users made their own decision about whether to join the intelligent care system. Informed patient consent forms were signed by the patients within the KidneyOnline app. All app users between January 2017 and January 2021 were screened according to the recruitment criteria, and eligible patients were grouped according to their original choice.
Patients were enrolled if they (1) were over 18 years old; (2) fulfilled the diagnosis of CKD (ie, eGFR <60 mL/min/1.73 m2 or eGFR < 90 mL/min/1.73 m2 but with albuminuria or hematuria for at least 3 months or as defined using other clinically indicated criteria); (3) uploaded 2 pieces of data upon registration with an interval of at least 3 months; (4) were free of dialysis, with a baseline eGFR >15mL/min/1.73 m2; and (5) were able and willing to provide informed consent. Patients were excluded if (1) they or their relatives were unable to use a smartphone, (2) they intended to start dialysis or have a kidney transplant within the next 3 months, (3) there was a lack of baseline or follow-up data, or (4) they refused to communicate with the health coach team. Study participants were followed for at least 3 months after recruitment until death or the start of renal replacement therapy.
The KidneyOnline intelligent patient care system was a nurse-led, patient-oriented collaborative management system as an adjunct to regular clinic visits for patients with CKD (
Participants in the KidneyOnline intelligent patient care group were able to experience at least 5 aspects of service provided by the system, which are detailed in
Patients in the conventional care group only had access to the health-related educational materials provided by the app. They could get professional suggestions from the nephrologists through regular clinical visits, but they did not have access to the service provided by AI and the health coach team.
Framework of KidneyOnline intelligent care system. AI: artificial intelligence; KDIGO: Kidney Disease: Improving Global Outcomes (clinical practice guidelines).
Interpretation of disease condition and corresponding guidance. The intelligent care system helped explain patients’ diagnoses and medication prescriptions, analyze and interpret lab results, and remind patients about medication precautions. The system also provided guidance on lifestyle interventions, including diet, exercise, and sleep. Patients could receive artificial intelligence (AI)–generated personalized recipes based on their disease condition and food preferences, and they were able to make food diaries to see if the nutrition requirements were met.
Regular check-ups. The intelligent care system followed each patient regularly to check if the patient was in good condition, assure if the recent BP was well-controlled, evaluate whether the patient’s current medication was reasonable, and assess whether the patient’s dietary intake was reasonable.
Early warnings. Through the algorithm, the intelligent system was able to identify risks associated with lab results and certain medications and make early warnings to minimize these risks. If abnormal lab results requiring immediate attention were identified, the system would send a medical alert to both the patient and the health coach. The health coach would take the initiative to contact the patient and make suggestions, including reminding the patient about medications, providing guidance on diet, exercise, and other lifestyle aspects, and reminding the patient to consult with doctors.
Real-time question-and-answer fields empowered by knowledge graphs. We established a renal knowledge graph according to the Kidney Disease: Improving Global Outcomes (KDIGO) guidelines for chronic kidney disease (CKD), which provided intelligent search and human-computer dialogues to enhance the efficiency and professional competency of the health coach team. Empowered by the graphs, the health coaches promptly provided real-time answers to any questions the patients had (
Clinical reminders. According to the patient's overall condition, the intelligent system could generate a personalized follow-up plan, reminding the patient to go to the doctor at regular intervals, helping them sort out test items they need to finish, and checking follow-up results to confirm whether the follow-up visit was completed.
Artificial intelligence (AI)–generated personalized recipes, blood pressure (BP) monitoring. and personal checklist dashboards of the KidneyOnline app.
The foundation of this intelligent system was built upon the structurization of patients’ health-related data using deep learning optical character recognition (OCR). Conventionally, patients’ health data came from a variety of sources, including (1) patients’ self-reported signs and symptoms; (2) data from intelligent home devices such as sphygmomanometer, and (3) patients’ past medical history, clinical notes, drug prescriptions, lab results, pathological reports, imagological exams, and so on. In the KidneyOnline intelligent system, patients uploaded those data simply by taking photos, and the intelligent system extracted the data efficiently by utilizing deep learning OCR. Combined with manual verification, a structured database was built so that patients’ health-related data could be integrated and analyzed quantitatively.
The participants took photos of their medical records, laboratory test results, and clinical prescriptions and uploaded them on the mobile app. All electronic data and photographs were uploaded instantly to a secure, cloud-based server. To comply with security and privacy regulations, patients’ smartphones were password-protected, and data were encrypted. Only the researchers were able to access the data on a cloud storage platform.
The primary outcome of our study was the first occurrence of either a 30% decrease in eGFR or an incidence of ESKD. Secondary end points included changes in 24-hour proteinuria and changes in mean arterial pressure (MAP). eGFR results were calculated by using the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) formula. Blood pressure data were collected through the app based on home blood pressure measurements uploaded by the patients, using either mercury or an electronic sphygmomanometer.
This study was approved by the local research ethics board from Anqing Municipal Hospital (2022-033) and was conducted according to the Declaration of Helsinki.
The distributional properties of data were expressed as mean (SD) for continuous variables with a normal distribution or median (IQR) for variables with a skewed distribution. For continuous data, the independent or paired Student
To balance the confounding factors between the KidneyOnline intelligent patient care system group and the conventional CKD care group, 1:1 propensity score matching (PSM) was performed using nearest neighbor algorithms with a 0.9 caliper width of 0.02 pooled standard deviations. Matching was based on baseline characteristics of age, sex, BMI, baseline eGFR, MAP, and proteinuria levels.
For survival analysis, both the Kaplan-Meier method and Cox proportional hazards model were used. The Kaplan-Meier method was applied to evaluate the cumulative incidence of primary outcomes in both groups following PSM. Additionally, the Cox proportional hazards model was utilized to identify predictive factors associated with the outcomes. For matched data after PSM, the Cox proportional hazards model with gamma frailty was used. All missing information was treated as missing data without imputation. Subgroup analyses were performed after stratifying according to the median age (<33 years vs ≥33 years), sex, kidney function (eGFR <60 mL/min/1.73 m2 vs ≥60 mL/min/1.73 m2), and proteinuria (<1 g/24 h vs ≥1g/24 h, <3 g/24 h vs ≥3 g/24 h), as well as the median value of baseline MAP (<88.7 mmHg vs ≥88.7 mmHg). Statistical analyses were performed using Python and Lifelines, an open-access survival analysis library written in Python. A 2-sided
Between January 2017 and January 2021, 78,007 potentially eligible patients were screened. Among them, 2060 (2.6%) were eligible for our analysis according to the inclusion and exclusion criteria. There were 1600 (77.7%) patients in the KidneyOnline intelligent patient care system group and 460 (22.3%) patients in the conventional care group (
Flow diagram of patient enrollment for the analysis. eGFR: estimated glomerular filtration rate.
Baseline and follow-up characteristics of patients in the whole cohort.
Demographics | Total (N=2060) | Conventional care group (n=460) | KidneyOnline care group (n=1600) | ||||||
Age (years), mean (SD) | 35.6 (9.5) | 35.3 (9.6) | 35.8 (9.5) | .32 | |||||
|
|||||||||
|
Female | 1175 (57) | 248 (53.9) | 927 (57.9) | .12 | ||||
|
Male | 885 (43) | 212 (46.1) | 673 (42.1) | .12 | ||||
BMI (kg/m2), mean (SD) | 22.7 (4.2) | 23.2 (4.8) | 22.5 (4) | .004 | |||||
|
|||||||||
|
IgAb nephropathy or IgA vasculitis | 945 (45.9) | 182 (39.6) | 763 (47.7) | .002 | ||||
|
Membranous nephropathy | 185 (9) | 49 (10.7) | 136 (8.5) | .18 | ||||
|
Focal segmental glomerular sclerosis | 63 (3.1) | 13 (2.8) | 50 (3.1) | .86 | ||||
|
Hypertensive nephropathy | 53 (2.6) | 23 (5) | 30 (1.9) | <.001 | ||||
|
Diabetic nephropathy | 7 (0.3) | 1 (0.2) | 6 (0.4) | .95 | ||||
|
Other types of nephritis | 246 (11.9) | 50 (10.9) | 196 (12.2) | .47 | ||||
|
Unknown etiology | 559(27.1) | 141(30.7) | 418(26.1) | .062 | ||||
|
Kidney transplant recipient | 2 (0.1) | 1 (0.2) | 1 (0.1) | .93 | ||||
|
|||||||||
|
Baseline eGFRc (mL/min/1.73 m2), mean (SD) | 88.6 (31.3) | 91.5 (31.4) | 87.8 (31.3) | .024 | ||||
|
CKD stage 3-4, n (%) | 443 (21.5) | 88 (19.1) | 355 (22.2) | .18 | ||||
|
Baseline proteinuria (g/24 h), median (IQR) | 0.8 (0.3-1.7) | 0.9 (0.3-1.9) | 0.7 (0.3-1.6) | .033 | ||||
|
Baseline MAPd (mmHg), mean (SD) | 88.6 (10.9) | 89.5 (11.1) | 88.4 (10.8) | .051 | ||||
|
|||||||||
|
RASBe | 1539 (74.7) | 333 (72.4) | 1206 (75.4) | .19 | ||||
|
Steroids | 524 (25.4) | 127 (27.6) | 397 (24.8) | .23 | ||||
|
Immunosuppressants | 418 (20.3) | 94 (20.4) | 324 (20.3) | .93 |
aCKD: chronic kidney disease.
bIgA: immunoglobulin A.
ceGFR: estimated glomerular filtration rate.
dMAP: mean arterial pressure.
eRASB: renin-angiotensin-aldosterone system blocker.
After a mean follow-up of 18.1 (SD 9.5) months, the primary composite kidney outcome occurred in 121 (8%) participants in the KidneyOnline intelligent patient care system group and 33 (7%) in the conventional care group, with 86 (5%) versus 21 (5%) and 35 (2%) versus 12 (3%) participants reaching a 30% eGFR decrease and ESKD, respectively.
Multivariate Cox regression with stepwise procedures was used to analyze the risk factors of composite kidney outcome. After adjustments for age, gender, baseline eGFR, proteinuria, MAP, and use of RASB, steroids, and immunosuppressants, individuals in the KidneyOnline care group were less likely to progress to ESKD compared with the conventional group, with a hazard ratio of 0.375 (95% CI 0.221-0.638;
A 1:1 PSM analysis was performed to balance the selection bias between the 2 groups. A total of 902 patients with 451 (50%) patients in each group were successfully matched. In total, the average age was 35.5 (SD 9.4) years, and 54% (n=487) were female. The average BMI was 22.9 (SD 3.8) kg/m2. The baseline MAP was 89.1 (SD 10.8) mmHg, and the 24-hour proteinuria was 0.8 (IQR 0.3-1.8) g/24 h. The baseline eGFR of the 902 patients was 91.9 (SD 30.7) mL/min/1.73 m2 (
Baseline and follow-up characteristics of patients after propensity score matching (PSM).
Demographics | Total (N=902) | Conventional care group (n=451) | KidneyOnline care group (n=451) | ||
Age (years), mean (SD) | 35.5 (9.4) | 35.27 (9.6) | 35.67 (9.2) | .51 | |
|
|||||
|
Female | 487 (54) | 244 (54.1) | 243 (53.9) | >.99 |
|
Male | 415 (46) | 207 (45.9) | 208 (46.1) | >.99 |
BMI (kg/m2), mean (SD) | 22.9 (3.8) | 22.9 (4.0) | 23.0 (3.6) | .90 | |
|
|||||
|
IgAb nephropathy or IgA vasculitis | 386 (42.8) | 181 (40.1) | 205 (45.5) | .13 |
|
Membranous nephropathy | 95 (10.5) | 45 (10) | 50 (11.1) | .64 |
|
Focal segmental glomerular sclerosis | 23 (2.5) | 12 (2.7) | 11 (2.4) | >.99 |
|
Hypertensive nephropathy | 29 (3.2) | 22 (4.9) | 7 (1.6) | .006 |
|
Diabetic nephropathy | 4 (0.4) | 1 (0.2) | 3 (0.7) | .63 |
|
Other types of nephritis | 99 (11) | 48 (10.6) | 51 (11.3) | .84 |
|
Unknown etiology | 265 (29.4) | 141 (31.3) | 124 (27.5) | .24 |
|
Kidney transplant recipient | 1 (0.1) | 1 (0.2) | 0 (0) | >.99 |
|
|||||
|
Baseline eGFRc (mL/min/1.73 m2), mean (SD) | 91.9 (30.7) | 91.5 (31.3) | 92.3 (30.2) | .63 |
|
CKD stage 3-4, n (%) | 163 (18.1) | 86 (19.1) | 77 (17.1) | .49 |
|
Baseline proteinuria (g/24 h), median (IQR) | 0.8 (0.3-1.8) | 0.8 (0.3-1.8) | 0.8 (0.3-1.6) | .29 |
|
Baseline MAPd (mmHg), mean (SD) | 89.1 (10.8) | 89.3 (11.0) | 88.9 (10.5) | .56 |
|
|||||
|
RASBe | 672 (74.5) | 327 (72.5) | 345 (76.5) | .20 |
|
Steroids | 232 (25.7) | 123 (27.3) | 109 (24.2) | .32 |
|
Immunosuppressants | 197 (21.8) | 93 (20.6) | 104 (23.1) | .41 |
aCKD: chronic kidney disease.
bIgA: immunoglobulin A.
ceGFR: estimated glomerular filtration rate.
dMAP: mean arterial pressure.
eRASB: renin-angiotensin-aldosterone system blocker.
After a mean follow-up period of 15.8 (SD 9.5) months, the primary composite kidney outcome occurred in 28 (6%) participants in the KidneyOnline intelligent patient care system group and in 32 (7%) in the conventional care group, with a hazard ratio (HR) of 0.391 (95% CI 0.231-0.661;
Kaplan-Meier curve of primary composite kidney outcomes.
Since the KidneyOnline intelligent patient care system showed different effects in different MAP groups, we further analyzed the MAP during follow-up between the KidneyOnline group and the conventional CKD care group. The baseline MAP values were similar between the 2 groups. The MAP significantly decreased from 88.9 (SD 10.5) mmHg at baseline to 85.6 (SD 7.9) mmHg at 6 months (
The mean level of 24-hour proteinuria in the KidneyOnline intelligent patient care system group was not significantly different from the conventional CKD care group at baseline (1.4 vs 1.5 g/24 h,
In this study, we described a nurse-led, smartphone-based patient-oriented system designed to help disease management in patients living with stages 1 to 4 of CKD. Our findings demonstrate that this intelligent care system was associated with better blood pressure control and a reduced risk of kidney failure. This provided a novel strategy for promoting a healthy lifestyle and improving kidney prognosis in patients with CKD, regardless of their scheduled consultations.
The incidence and prevalence of CKD have been persistently increasing because of the aging population, who experience a high incidence of metabolic disorders such as hypertension, diabetes, and obesity [
Our study demonstrated that the KidneyOnline care system reduced the risk of composite kidney end points, irrespective of age, baseline eGFR, and proteinuria. This effect was multifactorial. One of the main underlying reasons for this effect was the well-controlled blood pressure, which was significantly lower at 6 months compared with that at the time of enrollment in the KidneyOnline care group, and it remained at a lower level throughout the follow-up period. In China, the prevalence of hypertension is higher in patients with CKD compared with the general population, but awareness of hypertension in patients with CKD was reported to be 80.7% in 2018[
Interestingly, there was an improvement in proteinuria in both the KidneyOnline intelligent care group and the conventional care group. This may be attributed to the KidneyOnline app's educational materials on the benefits of a low-salt diet. In a typical Chinese diet, the salt intake is often much higher than recommended. The KidneyOnline app enabled patients from both groups to access free educational materials on the importance of a low-salt diet, as well as tips on limiting salt intake. Moreover, the KidneyOnline app offered a variety of low-salt diet recipes, which might have helped increase patients’ compliance with a low-salt diet. Additionally, there was a high percentage of patients with glomerulonephritis, with 25.7% (n=232) being treated with steroids. The use of steroids and immunosuppressants is another important factor leading to the improvement in proteinuria.
The rate of smartphone ownership has exploded in China during the last 10 years. Telehealth apps provide new opportunities to enhance self-management, behavior change, and medication adherence not only for people living in urban areas but also for those in rural areas. As Duan et al [
To our knowledge, this study was the first to assess the efficacy of a smartphone-based, nurse-led, patient-oriented management system for patients with CKD across China using hard kidney end points. The 4-year observational data from the real world demonstrated the efficacy of this newly developed system. Inevitably, our study had several limitations. First, there were over 70,000 user registrations on our app. Only 2060 patients’ records fulfilled our criteria and were eligible for analysis. This could have resulted in selection bias in both groups. Patients with a stronger sense of self-management and motivation were more likely to be included in this study. Second, patients who had severe symptoms or advanced pathological lesions were less likely to choose our patient care system; instead, they received treatments in large hospitals on their first visits. Moreover, patients with minor renal impairment could potentially have a lower probability of joining the KidneyOnline care system. In addition, due to the cost associated with the KidneyOnline care system, patients who joined the system probably had a higher economic status compared with patients in the control group. Nonetheless, we have shown that patients in the KidneyOnline care group experienced better prognoses in terms of composite renal outcomes, irrespective of their age, baseline eGFR, and proteinuria. More data are needed to illustrate the efficacy of our mode of care in those with more impaired kidney function, as well as those with very slight renal impairment. Third, we did not analyze other possible end points related to the utilization of the KidneyOnline care system, such as cardiovascular comorbidities and hospitalization frequencies, which would help us better evaluate our mode of management for patients with CKD. Additionally, the participants were young and normotensive; there was a high prevalence of glomerulonephritis but a limited number of diabetes cases, and these factors may have limited the feasibility of the KidneyOnline care system for patients with CKD caused by diabetic nephropathy. It is also important to note that our study was conducted in China; because the education and economic levels of patients with CKD vary across countries, a new mode of patient care based on mobile phones in other countries is anticipated.
We developed a smartphone-based, nurse-led, patient-oriented management system to facilitate health care for patients with nondiabetic CKD in China. Through multifaceted patient care, our mode of patient management was associated with a reduced risk of composite kidney outcomes. This strengthened the evidence of telehealth interventions to promote kidney health and long-term management for patients with CKD.
Demonstration of KidneyOnline dashboard for patients and health coach team.
Multivariate Cox regression analysis of composite kidney outcome (before patient matching). eGFR: estimated glomerular filtration rate; MAP: mean arterial pressure; RASB: renin-angiotensin-aldosterone system blocker.
Forest plot of the comparison between the 2 different modes of care in different subgroups (after patient matching). eGFR: estimated glomerular filtration rate; MAP: mean arterial pressure.
Changes in mean arterial pressure (MAP) during follow-up in the whole population and subgroups.
Changes in 24-hour proteinuria in the 2 matched groups of patients after propensity score matching (PSM) during follow-up.
artificial intelligence
chronic kidney disease
Chronic Kidney Disease Epidemiology Collaboration
estimated glomerular filtration rate
end-stage kidney disease
hazard ratio
Kidney Disease: Improving Global Outcomes
mean arterial pressure
National Kidney Foundation's Kidney Disease Outcomes Quality Initiative
optical character recognition
propensity score matching
renin-angiotensin-aldosterone system blocker
We are grateful to all the patients and controls for their participation in this study.
The data sets analyzed in this study are not publicly available due to business confidentiality reasons but are available from the corresponding author upon reasonable request.
Authors JW, TZ, XC, SX, and FH are employed by the Beijing Kidney Health Technology Co Ltd. The remaining authors have no conflicts of interest to declare.