Clinical Predictors and Algorithm for the Genetic Diagnosis of Pheochromocytoma Patients

Pheochromocytoma is a tumor of the paraganglial system. The nomenclature is not used consistently, and despite the limitation of the term to adrenal medullary tumors by the WHO (1), we, and others in this field, prefer to use it also for extra-adrenal retroperitoneal, pelvic and thoracic paraganglial tumors, because virtually all of them are vasoactive due to release of catecholamines in contrast to head and neck paragangliomas (HNP; refs. 2–4). We and others have shown that approximately one third of pheochromocytoma patients carry a predisposing germline mutation in one of six susceptibility genes (2, 5–9). The genes cause distinct clinical syndromes: von Hippel-Lindau disease (VHL; caused by germline mutations in the VHL tumor suppressor gene), multiple endocrine neoplasia type 2 (MEN 2; RET), paraganglioma syndromes type 1 (SDHD), type 3 (SDHC), and type 4 (SDHB) and type 1 neurofibromatosis (NF 1; due to mutations of the NF1 gene).

Hereditary HNPs are nearly exclusively associated with germline mutations of SDHB, SDHC, and SDHD (10, 11). In contrast, pheochromocytomas are frequently associated with germline mutations of five of the six genes, where only SDHC is rarely involved (12, 13). The molecular genetic classification of pheochromocytoma patients is an internationally accepted medical tool for molecular diagnosis, neoplasia risk assessment, genetic counseling, and clinical management.

When a single presentation, in this case, pheochromocytoma, can be attributed to any of several susceptibility genes, the cost of molecular genetic analysis becomes a factor in health care. For purposes of this study, we hypothesized that commonly obtained demographic and clinical features could guide us in prioritizing gene testing and hence, prove to be a cost reduction strategy in a diagnostic process. To address this hypothesis, we used our large European-American Pheochromocytoma Registry as our platform on which we systematically built a robust mathematical model to characterize clinical predictors that can be algorithmically used to prioritize whether genetic susceptibility should be considered, and if so, then which of the five genes is the most likely to be mutated.

We included index cases from the European-American Pheochromocytoma Registry who presented clinically with pheochromocytoma at the time of registration. In the situation where several subjects from one family were affected, only the index case of the family was used for purposes of this study.

Patients who developed pheochromocytoma after molecular-genetic testing was done were excluded. In addition, we excluded completely the many families in the Blackforest region in Germany, who mostly are unaware of being related to each other, but who carry an identical VHL mutation, almost certainly due to a founder effect (14).

We collected demographic and clinical data including gender, age at manifestation, tumor location, number of tumors, biology (benign versus malignant), family history for paraganglial tumors, extraparaganglial tumors in the index case, and for other tumors including thyroid, kidney, and pancreatic cancers as well as benign eye and central nervous system hemangioblastomas, and skin neurofibromas, which are the component phenotypic features suggestive of VHL, MEN 2, and NF 1.

Based on clinical retrospective data and family history, we defined as syndromic cases all patients fulfilling the clinical criteria for the diagnosis of NF 1, VHL, and MEN 2 syndromes. All other cases were defined a priori as nonsyndromic. Malignancy was defined as presence of documented distant or local-regional lymph nodal metastasis (1, 3).

Each registrant donated 10 mL EDTA- or ACD-anticoagulated blood, which was processed and banked for molecular genetic analyses.

Genomic DNA was extracted from peripheral blood leukocytes using standard procedures. We performed mutation analysis of the exonic and flanking intronic regions, including splice sites, mutations for all exons of the VHL (NM_000551.2), SDHB (NM_003000.2), SDHC (NM_003001), and SDHD (NM_003002.2) genes, and for exons 10,11, 13, and 16 of the RET (NM_020975.4) gene as described (2, 15). NF 1 was diagnosed clinically, because our systematic molecular genetic studies for mutations of the NF1 gene revealed mutations only in subjects who also have skin manifestations classic for NF 1 (5, 16).

In addition to PCR-based mutation scanning, this study also includes analyses for large deletions or rearrangements involving VHL, SDHB, SDHC, and SDHD, because these have been described by several groups (17–20), but not of the RET gene, because large deletions of this gene have not been described for MEN 2. For detection of large deletions/rearrangements, we performed multiplex genomic qPCR, as described (18). In parallel, we also used commercially available kits for multiplex ligation–dependent probe amplification for VHL, SDHB, SDHC, and SDHD (MRC-Holland). To confirm the deletions affecting either one PCR product (multiplex PCR) and/or one probe-ligation product (multiplex ligation–dependent probe amplification), additional multiplex PCR tests were designed. As controls, we used peripheral blood–derived DNA of 1,300 healthy consenting blood donors. These control subjects live in the same geographic areas and are of the same ethnicity as our registered pheochromocytoma patients.

Because we were seeking an unbiased statistical algorithm for prioritizing gene testing, we excluded from further analysis all syndromic cases, in particular all NF 1 cases and VHL and RET mutation carriers who also fulfilled either the clinical criteria for the diagnosis of VHL and MEN 2, respectively, and/or have positive family histories for either of these two syndromes. Furthermore, because SDHC mutations have been identified in only two cases, they have also been excluded from the analysis.

As a first step, multiple logistic regression analysis was done to identify predictors of a germline mutation in any gene. Seven variables were analyzed: age, gender, tumor number, tumor biology (benign versus malignant), tumor location (adrenal, extra-adrenal, and concomitant adrenal with extra-adrenal), previous HNP, and family history for paraganglial tumors. Recursive partitioning analysis was used to determine the age cutoff that best predicted any germline mutation.

As a secondary assessment of important predictors of any mutation, 1,000 bootstrap samples of size 989 were selected with replacement from the original set of patients. Stepwise logistic regression analysis using a variable entry criterion of P value of ≤0.10 and a variable retention criterion of P value of ≤0.05 was done on these bootstrap samples. Variables occurring with >50% frequency in these 1,000 analyses were considered to be important predictors.

Both the single sample analysis and the bootstrap validation identified four significant predictors of any mutation: age ≤45 y, multiple tumors, extra-adrenal location, and presence of HNP. Although malignant tumors and family history were not significant, they were included because of the importance of mutation screening in such patients for prognostic assessment (4, 21).

Six predictors were used to develop a screening algorithm to identify which patients should be genetically tested for mutations. If any of the six predictors is present, the patient should be tested for mutations, and if none are present, the patient should not be tested. Performance indicators for this strategy were examined in the sample of 989 patients. These indicators included the C-statistics for the multiple logistic regression model, and the sensitivity, specificity, and positive and negative predictive value of the screening algorithm. The major performance indicators evaluated in creating our model were the number of missed cases and the cost reduction.

As a second step, simple frequency counts of specific mutation types were examined among the subset of patients with at least one of the six predictors to establish the order in which the genes should be tested. The proposed screening algorithm is to test for genetic mutations in this order among the patients with at least one of the six predictors. The cost of this testing strategy was calculated in the sample of 989 patients, and was compared with the cost of testing all patients in the order in which the mutations occurred in these 989 patients (SDHB, VHL, RET, SDHD). For both approaches, costs were calculated per patient and rounded to the nearest whole dollar.

The performance of the six-factor logistic regression model, the predictive ability of the proposed screening algorithm, and costs were assessed in the 1,000 bootstrap samples. Results are summarized as the median, minimum, maximum, and 95% confidence interval. The confidence interval was calculated as the 2.5th percentile to the 97.5th percentile in the 1,000 samples. Analyses were done using SYSTAT v.10 software (SPSS, Inc.) or SAS software (SAS Institute, Inc.).

The U.S.-based costs of clinical mutation testing of each gene were obtained by averaging costs posted by GeneDx²⁵

The project was approved by the ethics committees/Institutional Review Boards for Human Subjects' Protection of the respective institutions. Participants provided written informed consent in accordance with the accepted standards for each respective country.

In total, 1,149 index cases presenting with symptomatic pheochromocytoma have been registered through December 2008. The median age-at-diagnosis was 42.3 years (5-83 years). The male/female ratio was 493:656. Of the 1,149 index cases, 91 (7.9%) had malignant disease, 243 (21.1%) had multiple pheochromocytomas, and 25 (2.2%) had a previous HNP presentation. The majority of index cases presented with unilateral adrenal tumors [816 (71.0%)], followed by bilateral adrenal [149 (13.0%)], extra-adrenal abdominal or pelvic [117 (10.2%)], and extra-adrenal thoracic tumor location [22 (1.9%)]. Concomitant presence of adrenal and extra-adrenal tumors was observed in 45 cases (3.9%).

The mutation frequency in our registry, considering all patients (syndromic and nonsyndromic) was 30.1% (346 of 1,149). Of these 346 cases, 43 (12.4%) had clinical evidence of NF 1. VHL germline mutations were detected in 120 (34.7%), RET in 80 (23.1%), SDHB in 73 (21.1%), and SDHD in 28 (8.1%) cases. SDHC germline mutations were detected in two cases, one of them, an intragenic mutation, previously published (12). Among mutation carriers, 18 were found to carry a large deletion or rearrangement: 12 involving SDHB, 4 VHL, and 1 each SDHD and SDHC. The SDHC deletion involved exons 5 to 6 of the gene in a 56-year-old male patient with a single extra-adrenal thoracic pheochromocytoma without family history for paraganglial tumors. The two SDHC carriers were excluded form further analysis.

After exclusion of the syndromic cases, in particular all the NF 1 cases, as well as 49 RET and 65 VHL mutation carriers with either personal or family histories of MEN 2 or VHL, respectively, a total of 989 eligible patients were available for further study. The mutation frequency of the remaining cases was 18.9% (187 of 989) with the majority affecting SDHB (73 of 187), followed by VHL (55 of 187), RET (31 of 187), and SDHD (28 of 187). The clinical and demographic features of the eligible cases after exclusion of these patients are represented in Table 1. From the 187 mutation carriers, 9.1% (17 of 187) had a positive family history for paraganglial tumors only.

The optimal age cutoff identified by recursive partitioning analysis was 42 years. To obtain a practical cutoff for clinical use, age was rounded to 40 and 45 years and both were assessed; age 45 years was chosen because it resulted in better identification of mutation-positive cases than age 40 years (data not shown). Seven variables were significantly associated with any mutation in univariable analysis, but only four were significant in multivariable analysis: age of ≤45 years, multiple pheochromocytoma/paraganglioma, extra-adrenal location, and previous HNP. This finding was confirmed by bootstrap analysis (Table 2).

Although malignant tumors and family history were not significant in the multivariable logistic regression analysis, they were included in the algorithm for genetic testing because of the importance of mutation screening in such patients for prognostic risk assessment (Table 3; refs. 4, 21). By preselecting the cases with at least one of these six predictors, 342 (34.6%) would be excluded from genetic testing. By doing so, 8 (4.3%) of 187 mutation carriers would be missed (Table 4).

Further, from data obtained by simple frequency tabulation, we established the order in which the genes should be tested (Fig. 1). This assessment indicated that patients with previous HNP were most likely to have SDHD, followed by SDHB, then RET, and least likely to have VHL mutations. Among patients with no prior HNP and a single extra-adrenal tumor, the order was SDHB as the most likely gene to be involved, followed by VHL, SDHD, and then RET. Among patients with no prior HNP and either multiple pheochromocytoma or a single tumor in a location other than extra-adrenal, the order was VHL as the most likely gene, followed by RET, SDHB, and SDHD as the least likely. As the major consequence, the screening algorithm as deduced is shown in Fig. 1.

If we chose to screen all pheochromocytoma-susceptibility genes (excluding NF1 and SDHC) for all patients presenting with pheochromocytoma, the average cost is currently ∼$3,400 per patient tested. Screening all patients in the gene order proposed (approach 1; Table 3), we achieve an 8.0% cost reduction without missing any mutation carrier. If we preselect the cases to be tested and test them in the order proposed (approach 2; Table 3), we can achieve a 39.8% cost reduction relative to approach 1, and a 44.7% cost reduction relative to testing all patients, but we miss eight mutation carriers (0.8% of all patients). The robustness of our results has been confirmed by bootstrap assessment analysis as shown in Table 3.

Our study confirmed that ∼30% of index cases had evidence, clinical or genetic, of one of the six known pheochromocytoma-associated syndromes. This frequency is much lower (18.9%) if clinical or anamnestical evidence of a syndrome (NF 1, VHL, and MEN 2) are excluded.

At the current state of knowledge, consequences of genetic testing play an important role in medical management, such as timely diagnosis of asymptomatic tumor manifestations, accurate genetic counseling and predictive testing, institution of gene-specific clinical surveillance, and/or prophylactic surgery (4, 22). According to the general recommendation for genetic screening in cancer-associated syndromes, all pheochromocytoma patients should be tested (23). However, the availability of genetic counselors and laboratories experienced with this relatively uncommon tumor as well as the costs of genetic analysis of multiple genes suggest the importance of prioritizing these analyses.

The most straightforward manner to help identify heritable pheochromocytoma is to retrieve and analyze the personal medical and family histories of each patient. In our series, from a total of 346 mutation carriers, clinical evidence of a syndrome (NF 1, MEN 2, or VHL) was present in 45% of the cases, which was confirmed by genetic testing. In particular, as previously published, all the cases harboring NF1 germline mutations had clinical evidence of this syndrome, and therefore, routine testing of this gene in pheochromocytoma patients is not recommended (5, 16).

But this type of syndromic recognition is insufficient. Excluding all these clinically obvious syndromic cases, we still have ∼19% (187 of 989) who are germline mutation carriers and only 9% of the 187 had a positive family history for paraganglial tumors. It is this subset of patients without the traditional syndromic “red flags” that is the focus of this study. Currently, this subset would be offered mutation testing of SDHB, SDHD, VHL, and RET. Our data confirmed that SDHC gene mutations are extremely rare in pheochromocytoma patients, and whether testing for SDHC in pheochromocytoma presentations in the absence of PGL should be done at all is currently not known (12, 13). To date, several demographic and clinical features have been associated with the presence of a germline mutation, such as young age at diagnosis, multifocal disease, extra-adrenal location, and malignancy (2, 7, 8, 21, 24). Thus, we took advantage of easily obtainable demographic and clinical features to come up with a robust algorithmic model to help predict who should be genetically tested and which gene to begin testing.

The major focus of our analysis was to obtain a model with the lowest number of missed genetic cases and highest cost reduction. Using the algorithm that would provide for identification of the highest proportion of individuals with mutations with the greatest cost reduction in the molecular diagnostic process, we would achieve a 44% cost reduction compared with testing all genes with the trade-off of missing eight individuals with mutations (0.8% of all 989 cases, 4.3% of 187 mutation carriers; Tables 2–4; Fig. 1). In contrast, if we used a model that picked up all cases with mutations, the overall cost reduction would only be 8%.

Currently, there are several proposed age cutoff points used to select cases for genetic screening (25–27). But, thus far, they have been based on frequency observations in small samples that were not statistically derived or confirmed in multivariable analysis. We have determined age ≤45 years to be an excellent cutoff in this large series of patients, and confirmed this finding in multivariable analysis.

From the eight cases (8 of 989) in whom the mutations were missed (Table 4), none developed further manifestations to date. Of interest is that in two of the three RET mutation carriers, clinical reinvestigation identified the presence of medullary thyroid carcinoma, which confirmed the presence of the MEN 2 syndrome. For one case, follow-up data are missing. These patients benefit immediately from genetic testing.

The costs can be further decreased, if testing for large VHL deletions is excluded. Our four cases with such VHL deletions had also clinical evidence of VHL and were excluded from this specific analysis.

For the SDHB and SDHD genes, however, large deletions or duplications were identified in 13% of the nonsyndromic pheochromocytoma patients with SDHB and SDHD mutations. Therefore, testing for large deletions of these genes should be done, after intraexonic mutations have been excluded.

When considering gene prioritization and cost reduction in this diagnostic process, special consideration is given to country-specific or ancestry-specific founder mutations. For example, 40% of all unselected male breast cancers and 10% of all female breast cancers diagnosed in Iceland are accounted for by a single mutation in BRCA2, 999del5 (28, 29). Similarly, several mutations in the SDHB, SDHD, and VHL genes have been observed to occur in higher frequencies in some restricted geographic areas and most likely represent founder mutations (14, 30–34). For patients coming from these areas, genetic testing for the region-specific founder mutation should be offered first in the absence of obvious syndromic features. If negative, then our algorithm should be followed.

Other genetic differences (e.g., variants) among distinct ethnicities and ancestries that might influence a patient's phenotype, and thus our algorithm, are currently not known, and therefore, our algorithm should be considered valid for all patients regardless of ethnicity. As data accumulate in this regard, our algorithm should be reevaluated and updated accordingly.

For purposes of genetic counselling, it should be noted that other mutation types or other as yet not identified pheochromocytoma-predisposing genes might exist (35, 36). In fact, in our series, 25 of 42 index cases with family histories of paraganglial tumors, no germline mutation in the known genes has been identified. Nonetheless, our algorithm can be easily recalculated after new gene or genes are identified.

With exponential advances in genetic and genomic technologies, many heritable disorders are being found to have several to multiple susceptibility genes. Importantly, subsets of these disorders present in a manner seemingly similar to that of sporadic disease. In the ideal world, these individuals should be offered genetic testing for all known susceptibility genes for that particular disease. However, this becomes impractical as health care costs soar. We provide here an algorithm that will help identify who should be offered genetic testing for nonsyndromic pheochromocytoma presentations and subsequently help prioritize which gene to begin testing based on modeled demographic and clinical features.

Study Group Members:

Germany: Alexander, Kappel-Grafenhausen; Allolio, Würzburg; Anlauf, Bremerhaven; Arlt, Würzburg; Arnold, Marburg; Bachmann, Arnsberg; Baumeister, München; Behrend, Freiburg; Bertram, Osnabrück; Beyer, Oberhausen; Beyersdorf, Deggendorf; Bergmann, Erlangen; Bierkamp, Bocholt; Bloch, Lübeck; Boehncke, Frankfurt; Bojunga, Frankfurt; Bosse, Stuttgart; Bootz, Bonn; Borany, Kleeve; Borch, Münster; Bornstein, Dresden; Brauckhoff, Halle; Brietzke, Schwerin; Büchler, Heidelberg; Callies, Hannover; Caspari, Bonn; Cupisty, Düsseldorf; Demtröder, Dortmund; Desiderato, Kappel-Grafenhausen; Dickert, Bodenwöhr; Diefenbach,/Roos, Beselich-Obertiefenbach; Dietz, Essen; Dörr, Erlangen; Donauer, Freiburg; Dost, Giessen; Dralle, Halle; Dupuis, Karlsruhe; Ebbinghaus, Bremen; Egli, Mönchengladbach; Eisenhofer, Dresden; Engert, Bochum; Fade, Homburg; Falk, Karlsruhe; Fassbinder, Fulda; Feldkamp, Bielefeld; Finke, Berlin; Fischer, Heidelberg; Fischer, Trier; Fritze, Rheine; Frilling, Hamburg; Furtwängler, Freiburg; Gal, Garbrecht, München; Hamburg; Gerke, Freiburg; Glien, Bonn; Götte, Bad Hersfeld; Goretzki, Düsseldorf; Graubner, München; Grotz, Essen; Haag, Frankfurt; Hagen, Oberhausen; Hahner, Würzburg; Harsch, Erlangen; Hartung, Düsseldorf; Hauch, Hamburg; Happ, Frankfurt; Hauschild, Lübeck; Hecht, Leipzig; Hector, Hamburg; Hehmann, Stuttgart; Hehr, Regensburg; Heidemann, Augsburg; Heidemann, Stuttgart; Heins, Osnabrück; Hempel, München; Heppner, Köln; Hamburg; Herbach; Karlsruhe; Herlan, Bötzingen; Hetzer, Berlin; Hirsch, Offenburg; Hofmeier, Windsbach; Hofmockel, Würselen; Holbeck, Solingen; Hopt, Freiburg; Host, Münster; Hudelmayer, Pforzheim; Isermann, Heidelberg; Iwen, Lübeck; Jacobs, Osnabrück; Jarzab, Gleiwitz; Jocham, München; Jürgens, Münster; Judy, Giessen; Junginger, Frankfurt; Jurdi, Niederfischbach; Justl, Würzburg; Kabisch, Hamburg; Karaa, Beeskow; Kavelage, Hamburg; Kindler, Würselen; Kisters, Herne; Kirste, Freiburg; Klein-Franke, Göttingen; Klingebiel, Frankfurt; Koch, Leipzig; Köster, Hamburg; Kornely, Duisburg; Krämer, Warstein; Kremens, Essen; Krone, Köln; Krumme, Wiesbaden; Kühn, Freiburg; Kusterer, Mannheim; Lampe, Magdeburg; Lamster, Erfurt; Lederbogen, Essen; Lee, München; Leisner, Kirchheim; Lindemann, Ettlingen; Lohmann, Dresden; Löβner, Magdeburg; Lohöfener, Berlin; Lorenz, Halle; Lückhoff, Landshut; Luft, Berlin; Macazhavy, Düsseldorf; Malyar, Münster; Mann, Essen; Meyer, Erfurt; Meyer, Frankfurt; Mitter, Essen; Mittler, Magdeburg; Möller, Essen; Mönig, Kiel; Nawroth, Heidelberg; Netzer, Köln; Neumann, Berlin; Niebling, Titisee; Niedermeyer, Düsseldorf; Niemeyer, Freiburg; Nies, Osnabrück; Noah, Villingen; Normann, Düsseldorf; Opitz, Schwetzingen; Otto, Freigericht; Pätsch, Hannover; Pasedag, Regensburg; Pavel, Erlangen; Pavenstädt, Münster; Peter, Freiburg; Petzold, Dresden; Pfäffle, Leipzig; Pichl, Nürnberg; Pielken, Oberhausen; Plingen, Düsseldorf; Ploner, Stuttgart; Pohl, Stuttgart; Quante, Düsseldorf; Racasan, Frankfurt; Reichhardt, Leipzig; Reincke, München; Reiβenweber, Bamberg; Reschke, Magdeburg; Reuter, Cuxhaven; Rieg, Freiburg; Riel, Erlangen; Ritter, Münster; Röher, Düsseldorf; Rohrer, Homburg; Roth, Bonn; Rudolph, Risum-Lindholm; Rudolph, Bad Berka; Rückauer, Freiburg; Schäfer, Magdeburg; Schäfer, Pfarrkirchen; Schellong, Erlangen; Scherbaum, Düsseldorf; Schlottmann, Unna; Schmidt, München; Schmid, München; Schmitz, Bonn; Schöninger, München; Schogohl, München; Scholz, Mainz; Schott, Düsseldorf; Schulze, Freiburg; Schumann, Köln; Schumm-Dräger, München; Schweinfest, Emershausen; Seppel, Mönchengladbach; Seufert, Freiburg; Siegler, Frankfurt; Sievers, Bederkesa; Simon, Düsseldorf; Sieverts, Heidelberg; Slisko, München; Sommer, Osnabrück; Speicher, München; Stahl, Lörrach; Stephan, Karlsruhe; Stoll, Offenburg; Tan, Essen; Trupka, Starnberg; Unger, Essen; Vogelsang, München; Wahl, Frankfurt; Walter, Limburg; Weber, Mainz; Weber, Schaffhausen; Weber, Wickede; Wehinger, Kassel; Weisbrod, Hasloch; Wenzel, Essen; Wiegand, München; Wiiwer, Waldeck; Willenberg, Düsseldorf; Wilms, Bonn; Yüce, Essen; Zimmerhackl, Freiburg; Zimmermann, Chemnitz; Zumkeller, Halle; zur Mühlen, Hannover;

Austria: Fauth, Innsbruck

Canada: Winquist, Ontario

Chile: Bello, Candia

England: MacGregor, London; Bockenhauer, London; Schulte, London;

Finland: Aittomäki, Helsinki; Paronen, Helsinki; Valimäki, Helsinki; Virkamäki, Helsinki;

France: Brunaud, Nancy; Klein, Nancy; Limacher, Straβbourg; Pigny, Lille; Weryha, Nancy;

Greece: Zamboulis, Thessaloniki;

Israel: Regev, Assaf Haroteh; Glase, Jerusalem;

Italy: Falcioni, Piacenza; Riegler, Bozen; Schiavi, Padova, Stanzial, Bozen;

Norway: Heimdal, Oslo, Laurent, Oslo

Romania: Milos, Timisoara

Spain: Fuentes, Pamplona; Rovia, Madrid;

Sweden: Ramme, Stockholm; Wallin, Stockholm

Switzerland: Angst, Aarau; Mohaupt, Bern; Müller, Basel; Pichert, Zürich; Preiswerk, Zürich; Sanssonnens, Bern; Stettler, Bern; Sulzer, Zürich; Unger, Lausanne; Walter, Basel;

Turkey: Huseyn, Ankara

Ukraine: Kollkh, Kiev; Yaremchuk, Kiev;

USA: Uckun-Kitapci, Chapel-Hill

No author had any financial and personal relationships with other people or organizations that could inappropriately influence or bias this work.

We thank all patients who contributed to this work, Aurelia Winter (study assistant) for her technical support, and Mary Buchta, Gani Berisha, and Tobias Blüm (technicians) working at the Department of Nephrology, Section of Preventive Medicine, Albert-Ludwigs University, Freiburg, Germany.