Incidence and Relapse of Idiopathic Nephrotic Syndrome: Meta-analysis

This systematic review and meta-analysis was written according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guideline and the Meta-analyses of Observational Studies in Epidemiology guideline for intervention studies and observational studies, respectively.^20,21  This review has been registered on the International Prospective Register of Systematic Reviews (identifier CRD42020154571).

A comprehensive search was performed in the Medline and Embase databases. No restrictions to language, year of publication, or study design were applied. Databases were searched until December 2020. Briefly, the following Medical Subject Headings terms were included in the final search: “idiopathic nephrotic syndrome OR nephrosis” AND “children” AND “epidemiology OR incidence” AND “steroids.” See the Supplemental Information for the complete search strategy (Appendix 1). Additionally, references were checked for potentially eligible studies.

Rayyan (rayyan.qcri.org), a free Web-based screening tool,²²  was used to screen for titles and abstracts. Two independent reviewers (F.V. and L.R.R.) selected studies for full-text screening, after which final inclusions for analysis were selected by using the following inclusion criteria: (1) first episode of INS (see Table 1 for definitions), (2) children aged 0 to 18 years, (3) incidence as new cases of INS per 100 000 children per year and/or the risk for relapse as the proportion of patients who relapsed after the first episode of SSNS (Table 1) or could otherwise be calculated, and (4) observational (both retro- and prospective) and intervention studies. Studies were excluded if the study did not contain original data, reported only patients with secondary or congenital forms of nephrotic syndrome, consisted of subgroups of INS (ie, biopsy-proven MCNS only, steroid-resistant nephrotic syndrome, or accompanied by acute kidney injury or other comorbidities), or reported duplicate cohorts or when no full-text publication was available. In the cases in which no full-text was available, the authors were contacted. Discrepancies between reviewers were solved through discussion first and, if left unsolved, by a third independent reviewer (A.H.M.B.).

Two independent reviewers assessed the methodologic quality of each study using a modified 11-item list based on the Strengthening the Reporting of Observational Studies in Epidemiology checklist (for observational studies),^23–25  the Cochrane Risk of Bias checklist (for randomized controlled trials [RCTs]),²⁶  or Risk Of Bias In Non-randomized Studies of Interventions (for nonrandomized studies).²⁷  High quality was defined as a quality score of ≥75%; low quality was defined as a quality score of <75%. A cutoff value of ≥75% was chosen to make sure that at least 8 items of each checklist need to be positively checked to be classified as high quality and is in line with previous systematic reviews by using quality scores.²⁸  Interrater variability was determined by using Cohen’s κ-coefficient. Publication bias for relapse studies was assessed graphically by using a funnel plot as well as Kendall’s τ rank correlation test and Egger’s regression test.

Data were collected by using a standardized data extraction form on Google Forms. After quality assessment, 2 reviewers extracted the following data from included studies independently: the first author’s name, year of publication, year of conduct, duration of the study, country of conduct and income level according to the World Development Indicators of the World Bank,²⁹  study design, study population (main inclusion criteria), sample size, total population at risk, overall incidence and incidence per ethnicity (white, Black, South Asian, Southeast Asian, East Asian, Middle Eastern and Northern African, and Latin American [people of Central and South American origin]), patient’s baseline characteristics (age at onset, sex distribution, and ethnicity), initial treatment (long or short course steroids and/or adjuvant therapy), steroid-responsiveness, overall risk for relapse (proportion of patients with SSNS who experienced at least 1 relapse), risk for relapse per duration of initial steroid therapy, mortality, and duration of follow-up.

Separate random-effects meta-analyses by using the DerSimonian–Laird method were used to calculate the overall pooled weighted estimates and their corresponding 95% prediction intervals (PIs) of incidence and risk for relapse of INS in children. Because Banh et al¹⁷  reported 3 separate cohorts in 3 different time periods, each cohort was individually analyzed. Given that the risk for relapse was not normally distributed and is an absolute measure, all analyses were performed after Freeman–Tukey double arcsine transformation. Transformation was not possible for incidence because, in the original studies, the researchers did not provide person-time data. For each study, the median year of conduct was calculated. To identify a change in the incidence and risk for relapse over time, a regression coefficient was calculated by using univariate meta-regression. Because substantial heterogeneity in the population among studies was expected, it was assumed that the included studies present a random sample of the general population and, therefore, a random-effects model was preferred. PIs were calculated as follows:

where $μ˘$ is the pooled weighted estimate of the random-effects model, $τ˘2$ is the SD of the between-study variation, and SE(⁠$μ˘$⁠) is the SE of the pooled weighted estimate. Statistical heterogeneity among studies was calculated by using I² statistics. Heterogeneity was considered unimportant, moderate, substantial, or considerable if I² was 0% to 40%, 30% to 60%, 50% to 90%, or 75% to 100%, respectively.³⁰ 

To explore potential sources of heterogeneity, several subgroup analyses for global region of conduct, ethnicity (incidence only), and steroid treatment duration (long [≥4 months] versus short [<4 months]; relapse only) were performed by using separate random-effects models. Univariate meta-regression models were used to identify potential sources of heterogeneity among included studies on the effect of incidence and risk for relapse. To test the robustness of the results, subsequent sensitivity analyses were conducted. Primary analyses for incidence and risk for relapse were repeated substituting a priori specified variables: income level, initial steroid response, steroid treatment duration, study design (prospective only), inclusion criteria (age <18 years or age ≤12 years; well-defined INS only), population at risk (≥500 000 children only) or sample size (≥50 or ≥100 patients only), and duration of follow-up (≥12 months or ≥24 months only). For risk for relapse, the meta-analysis was repeated by using generalized linear mixed models (GLMMs) with logit transformation of the data. Additionally, individual studies with potential influence on the heterogeneity were identified visually by using a Baujat plot, followed by using studentized residuals and a leave-one-out analysis. Studies with a residual ≥2.8 or ≤−2.8 were considered to be of influence.

A P value <.05 was considered statistically significant. The metafor package for R (version 3.6.1) was used for all statistical analyses.^31,32 

The initial search yielded 2573 studies after removal of duplicates (n = 677). After title and abstract screening, a total of 195 full-text articles were screened by 2 independent reviewers. Ultimately, 27 studies in which researchers reported on incidence and 54 studies in which researchers reported on risk for relapse were included, with 8 studies in which researchers reported both outcomes.^17,33–39  The most frequent reasons for exclusion were the following: not reporting the outcome (n = 35), no full-text available (n = 33), the study design (n = 24), and reporting on subgroups only (n = 17; Fig 1). In total, 15 666 patients were included in all studies. Sample sizes ranged from 18 to 4083,^40,41  with a median sample size of 114. All studies on incidence were observational studies or cohort studies, whereas 14 (30%) on relapse were RCTs. More than one-half of the studies were conducted as single center (57%), and two-thirds were of retrospective design (66%). Most studies took place in Europe (n = 27); 2 ISDKC studies were conducted on multiple continents.^42,43  The mean and median age of the study subjects ranged from 3.1 to 11.6 years and 2.50 to 12.00 years, respectively. Male subjects were more often affected than female subjects (mean: 64.3%). A median of 89.4% (range: 54.7%–100%) of the patients responded to initial steroid therapy (Table 2). A detailed overview of the study and patient characteristics of the included studies is provided in the Supplemental Table 5.

The overall quality of the included studies was high for both observational (n = 59; mean quality score: 78%; range: 55%–100%) and intervention studies (n = 14; mean: 77%, range: 56%–100%). Of the 2 nonrandomized controlled studies, 1 was scored as high quality (89%),⁴⁴  and 1 was scored as low quality (56%).⁴⁵  Although 24 (45%) of all studies were scored as low quality, 14 of these had a quality score between 70% and 75% (Supplemental Information 2). Interrater variability was moderate to strong, corresponding with a Cohen’s κ coefficient of 0.69 and 0.83 for observational and intervention studies, respectively. For studies reporting risk for relapse, there was no publication bias (Supplemental Fig 6).

The overall pooled weighted estimate of the incidence of first onset INS was 2.92 (95% PI: 0.00–6.51) new cases of INS per 100 000 children per year (Fig 2). There was considerable heterogeneity across studies (I² = 96.5%), of which 87.5% could be explained by differences in global regions. The incidence of INS was higher in Southeast Asia and East Asia (6.11; 95% PI: 4.81–7.42; I² = 79.3%), compared with Europe (2.15; 95% PI: 0.80–3.69; I² = 80.6%), North America (2.38; 95% PI: 1.43–3.34; I² = 45.0%), and Oceania (1.43; 95% PI: 0.54–2.32; I² = 50.3%; Fig 3A). A significant lower incidence was found in Western countries (2.18; 95% PI: 1.84–2.52; I² = 78.0%), compared with non-Western countries (6.07; 95% PI: 4.58–7.56; I² = 55.3%; P < .001) (Supplemental Fig 7). The incidence was the lowest in white children; however, all PIs overlapped (Table 3; Supplemental Fig 8). Meta-regression analysis revealed no significant change in the incidence between 1929 and 2011 (P = .39; Fig 4A).

The overall pooled weighted estimate of the risk for relapse of first onset INS was 71.9% (95% PI: 38.8–95.5), with considerable heterogeneity across studies (I² = 94.9%; Fig 5). Unlike the incidence data, no single important driver of heterogeneity could be identified. There was no significant difference between a long (65.0%; 95% PI: 29.3–93.3; I² = 94.5%) and short (72.2%; 95% PI: 35.6–97.2; I² = 94.7%) duration of initial steroid treatment (P = .16; Supplemental Fig 9). The risk for relapse ranged among global regions from 53.6% (95% PI: 0.00–100; I² = 97.3) in Latin America to 84.2% (95% PI: 76.2–90.8; I² = 51.5) in North America (Table 3; Fig 3). Western countries reported a significantly higher risk for relapse (80.9%; 95% PI: 53.0–98.1; I² = 92.9) than non-Western countries (63.2%; 95% PI: 31.6–89.7; I² = 94.3; P < .001; Supplemental Fig 7B). Data on the risk for relapse per ethnicity were not available; therefore, no separate analyses could be performed. Studies with a sample size <50 or <100 patients or a duration of follow-up <12 months would explain 19.2%, 13.8%, and 32.6% of the heterogeneity, respectively. Meta-regression analysis revealed a significant decrease in the risk for relapse between the earliest and latest median year of conduct (r² = −0.0030; P = .033; Fig 4B). Between 1945 and 2011, the risk for relapse decreased from 87.4% (95% PI: 53.9–100) to 66.2% (95% PI: 32.3–92.8), an absolute decrease of 21.2%.

Subsequent sensitivity analyses revealed that, for incidence, substituting the a priori defined variables had no effect on the overall estimate (Table 4). The study by Elzouki et al⁴⁶  was identified by using a Baujat plot as outlier for incidence, but this finding could not be confirmed by using its studentized residual (residual = 2.72) and the leave-one-out analysis (I² = 96.6). In contrast, the study by Kikunaga et al³⁶  was identified as a potential source of heterogeneity (residual = 3.36). By leaving out this study, heterogeneity would drop to 90.5% but would still be considerable. Furthermore, it had little influence on the overall estimate of the incidence (2.62). For risk for relapse, excluding studies with a small sample size or short follow-up had an effect on neither the overall estimate of the risk of relapse (Table 4) nor change over time. Of the other variables, none were of relevance. Of the individual studies, the study by Zhang et al⁴⁷  had a residual of −3.72 but an effect on neither the heterogeneity (I² = 93.2) nor overall estimate (72.9%). Repeating the meta-analysis for risk for relapse by using a GLMM with a logit transformation of the data led to a minimal change in the overall estimate (73.2%; 95% PI: 31.8–94.1; I² = 95.8%), also maintaining a significant decrease over time.

The results from this systematic review and meta-analysis reveal that INS has a low global incidence but a high risk for relapse. Whereas incidence was significantly lower in Western countries, children in non-Western countries have a lower risk for relapse. Incidence tended to be lower in white children, but this was not significant. No separate data on the risk for relapse per ethnicity was available. There was no indication that the incidence changed over time, yet a 21.2% decrease from 88.4% to 67.2% in the risk for relapse between 1945 and 2011 was observed.

The lower incidence in children from Western countries is most likely explained by differences in ethnicity. We found a lower but not significant incidence in white children. Ethnical differences in the onset of INS, most explicitly in children of South Asian origin, have also been observed by others.⁴⁸  In this study, the incidence of INS in children of South Asian descent was 3 to 4 times higher than in white children. The retrieved incidence rates for South Asian children are from subgroup analyses of studies conducted in Europe and North America because there were no incidence studies from South Asian countries. Therefore, we cannot exclude the idea that the global incidence is an underestimation because the reported incidence from Western countries is significantly lower. Although ethnical differences in the risk for relapse could not be separately analyzed, a significant difference between Western and non-Western countries exists. This difference might be explained by ethnical differences but may also be a result of factors that could be further explored, such as loss to follow-up. Heterogeneity in both groups was still considerable, resulting in overlapping PIs, and, therefore, this finding should be carefully interpreted.

For both the global incidence and risk for relapse, considerable heterogeneity among studies was identified. Statistical heterogeneity is a result of clinical and methodologic diversity. High degrees of heterogeneity may add uncertainty to the results but, also, reflect a more general population. Several study (methodologic diversity) and patient (clinical diversity) characteristics as potential sources of heterogeneity were explored. For incidence, study location explained most of its heterogeneity, but leaving global regions out of the meta-analysis had no effect on the heterogeneity of the overall estimate. However, including only Western or non-Western countries in the analysis decreased the degree of heterogeneity in both groups to 78.0% (substantial) and 55.3% (moderate), respectively. Because of these geographical differences, our global estimate of the incidence of INS may be underestimated because the majority of the studies were conducted in Europe (n = 10) and North America (n = 10), which have the lowest incidences. Another factor contributing to the heterogeneity is the potential inaccuracy of the census data used in some, mostly older studies or studies from middle- or lower-income countries.

The risk for relapse has been (partially) influenced by the sample size and duration of follow-up (methodologic diversity). Although there was a positive association between sample sizes larger than 50 or 100 patients and the risk for relapse, heterogeneity could only be partly explained by the sample sizes. Small sample sizes are prone to great variability, affecting the reliability of the results. Of the included studies with a small sample size, the risk for relapse ranged from 29.4% to 100%.^40,49 

As shown in various Kaplan–Meier curves for relapse-free survival, a first relapse generally occurs in the first 2 years after initial presentation.^50,51  Therefore, it is possible that in studies with short follow-up periods, researchers underestimated the risk for relapse. This was confirmed by a recalculation including only studies with a follow-up <12 months (P < .001) but not studies with a follow-up <24 months (P = .11). There was a decrease in the median duration of follow-up over time, but this had no impact on the overall estimate or decrease of risk for relapse over time.

Using the subsequent sensitivity analyses, we confirmed that the results of the meta-analyses are robust to the decisions made in the selection of studies. Because the Freeman–Tukey transformation is debated,⁵²  all analyses concerning the risk for relapse were repeated by using a GLMM with a logit transformation of the data. This had a minimal effect on the results but did not alter the conclusions. For risk for relapse, excluding studies on the basis of small sample size or short follow-up altered neither the degree of heterogeneity nor the overall estimate or change over time. Although Kikunaga et al³⁶  was identified as an individual study with a large influence on the heterogeneity, the study was not excluded from the analysis because it was considered one of the higher quality nationwide studies. This study was probably identified as being of influence because it had the largest population at risk. In conclusion, no studies were excluded from the primary meta-analyses and independent from the choice of model.

To facilitate the interpretation of the results and translation into clinical practice, PIs, rather than confidence intervals, were calculated. PIs are used to report which range of true estimates can be expected in 95% of future studies and which outcomes are to be seen in clinical practice.^53,54  Because PIs take between-study variation into account, higher degrees of heterogeneity will result in wider PIs. Consequently, PIs will generally provide wider intervals than confidence intervals but are clinically more relevant. The PIs of incidence indicate that the global incidence should lie between 0 and 6.5, but it is likely to be closer to 4 or 5. The risk for relapse for any patient is (on the basis of its PI) between 40% and 95%.

This study has several shortcomings. A number (n = 33) of potentially eligible publications was not available in full text and, therefore, data on the incidence and risk for relapse could have been missed. Additionally, limited- to no-incidence studies from Africa (n = 1), South Asia (n = 0) and Latin America (n = 0) were identified. Furthermore, because there was no limitation to year of publication, a wide range of definitions of INS have been used. In early studies, researchers included patients with “nephrosis,” having substantial but not quantified proteinuria and hypoalbuminemia in the presence of edema. Later, more clear definitions of INS were defined by using the ISKDC group and Kidney Disease Improving Global Outcomes (KDIGO) guidelines. Still, in several recent studies,^17,38  the researchers did not clearly report which definition was used, which could have contributed to the considerable heterogeneity. However, including only studies in which INS was well defined by the ISKDC or KDIGO criteria did not change the overall estimate. Furthermore, although almost all incidence studies were of retrospective design, potentially missing eligible INS patients, in several studies, researchers reported high response rates of surveys,^9,15,36  or searched national databases by using the International Classification of Diseases codes.^18,41,55,56 

The fact that the incidence of INS did not change over time and that higher incidences in South Asian children were reported in different but mainly Western countries suggests that genetic factors are likely to be of influence on the pathogenesis of INS. The role of genetic defects in the pathogenesis in SSNS is still largely unclear. This is in contrast to steroid-resistant nephrotic syndrome, in which >50 single genes are identified that can explain up to 30% of the cases.⁵⁷  Differences in the occurrence and severity of SSNS may be, in part, explained by HLA class I and II alleles. HLA class II is required for antigen presentation and activating T cells by production of surface proteins on immune cells. In several studies, researchers confirmed that risk loci are specifically located in the HLA-DQ and HLA-DR region.^58–64  However, transethnic genome-wide association study analysis revealed that this association is true for different ethnicities.⁶²  Only HLA-DRB1 seems to be more specific for the Japanese population.⁶⁰  Further research into the genetic causes of SSNS is needed to explain the variability between ethnicities.

To further decrease the risk for relapse after the first episode of INS, future research should be focused on therapies other than corticosteroid treatment only. Since the introduction of steroids, many studies were conducted in which researchers optimize corticosteroid treatment schedules. In earlier studies, researchers indicated that a longer duration and/or higher doses of steroids were superior to a short duration (<4 months). However, a series of more recent, well-conducted studies revealed no beneficial effect of long course steroid treatment.^51,65–67  Still, although relapses are less common, there is ample room for improvement. To treat relapses, patients often require long-term corticosteroid therapy, leading to short- (binge eating, behavioral changes, cushingoid features including stretch marks, and a higher risk for infections) and long-term (weight gain, hypertension, osteoporosis, risk for diabetes, and cataract) complications. Currently, 3 large, multicenter RCTs have been conducted to investigate the efficacy and safety of adding a second agent to the initial treatment of INS. Whereas the Levamisole as Adjuvant Therapy to Reduce Relapses of Nephrotic Syndrome study (the Netherlands and Belgium)⁶⁸  and the Nephrovir-3 study (France) are focused on levamisole versus placebo added to corticosteroids, the Initial Treatment of Steroid-Sensitive Idiopathic Nephrotic Syndrome in Children With Mycophenolate Mofetil Versus Prednisone: Protocol for a Randomised, Controlled, Multicentre Trial study⁶⁹  is focused on mycophenolate mofetil plus short-term prednisone versus prednisone alone. Both levamisole and mycophenolate mofetil are used as steroid-sparing drugs in the treatment of frequently relapsing and steroid-dependent INS, with similar effectiveness and an acceptable side effect profile. The results from these trials are to be expected within the coming years and will hopefully reveal promising results to prevent relapses of INS even further and, thereby, reduce the burden of the disease and its treatment.

INS in children has a low incidence in the general pediatric population that has remained stable over time. Higher incidences were found in non-Western countries, which are likely explained by ethnical differences. Although decreasing, the risk for relapse remains high. To prevent relapses of INS even further, large (international) multicenter RCTs are needed.

The LEARNS consortium is an interuniversity collaboration in the Netherlands that is established to perform a double-blind, placebo-controlled RCT on the efficacy of levamisole on relapses in children with a first episode of SSNS and study the basic mechanisms underlying the nephrotic syndrome and the mode of action of levamisole. Members of the consortium include (in alphabetical order): A.H.M. Bouts, principal investigator (Department of Pediatric Nephrology, Amsterdam UMC, location AMC, Amsterdam, the Netherlands), S. Florquin (Department of Pathology, Amsterdam UMC, location AMC, Amsterdam, the Netherlands), J.E. Guikema (Amsterdam UMC, location AMC, Amsterdam, the Netherlands), L. Haverman (Psychosocial Department, Amsterdam UMC, location AMC, Amsterdam, the Netherlands), L.P.W.J. van den Heuvel (Radboud Institute for Molecular Sciences, Radboud University Medical Center, Nijmegen, the Netherlands), E. Levtchenko (Department of Pediatric Nephrology, University Hospitals Leuven, Leuven, Belgium), R.A.A. Mathôt (Department of Hospital Pharmacy, Amsterdam UMC, location AMC, Amsterdam, the Netherlands), M.F. Schreuder (Department of Pediatric Nephrology, Radboud University Medical Center, Nijmegen, the Netherlands), B. Smeets (Radboud Institute for Molecular Sciences, Radboud University Medical Center, Nijmegen, the Netherlands), and J.A.E. van Wijk (Department of Pediatric Nephrology, Amsterdam UMC, location AMC, Amsterdam, the Netherlands).

We thank René Spijker from the Medical Library of the Amsterdam UMC, location AMC, for his help with the final search strategy as well as Mariska Leeflang from the Department of Clinical Epidemiology, Biostatistics, and Bioinformatics of the Amsterdam UMC, location AMC, for her help with the statistical analyses, specifically the meta-analyses, and Arthur Edridge from the Emma Children’s Hospital, Amsterdam for critically revising the article.

GLMM

generalized linear mixed model

INS

idiopathic nephrotic syndrome

ISKDC

International Study of Kidney Disease in Children

KDIGO

Kidney Disease Improving Global Outcomes

MCNS

minimal change nephrotic syndrome

PI

prediction interval

RCT

randomized controlled trial

SSNS

steroid-sensitive nephrotic syndrome