Understanding individual drug response variation
Kroonen, Marjolein
DOI:
10.33612/diss.127010643
IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document version below.
Document Version
Publisher's PDF, also known as Version of record
Publication date: 2020
Link to publication in University of Groningen/UMCG research database
Citation for published version (APA):
Kroonen, M. (2020). Understanding individual drug response variation: Pharmacokinetic analysis of diabetes trials. University of Groningen. https://doi.org/10.33612/diss.127010643
Copyright
Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).
Take-down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum.
Understanding individual
drug response variation
Pharmacokinetic analysis of diabetes trials
Door M.Y.A.M. Kroonen
Colophon
Financial support for the publication of this thesis by the University of
Groningen, University Medical Center Groningen, Graduate School of Medical Sciences (GSMS), Noord Negentig, Boehringer Ingelheim,
KNMP and Astra Zeneca are gratefully acknowledged. ISBN
ISBN: 978-94-034-2711-9 (Printed version) ISBN: 978-94-034-2712-6 (Digital version) Cover design
M.E.A.M. Kroonen Layout
M.Y.A.M. Kroonen Printed by
Ipskamp Drukkers, Enschede
© 2020, M.Y.A.M. Kroonen, Groningen, the Netherlands
All rights reserved. No part of this publication may be reproduced, copied, modified, stored in a retrieval system, or transmitted in any form without permission of the author.
Understanding individual
drug response variation
Pharmacokinetic analysis of diabetes trials
Proefschrift
ter verkrijging van de graad van doctor aan de Rijksuniversiteit
Groningen
op gezag van de
rector magnificus prof. dr. C. Wijmenga
en volgens besluit van het College voor Promoties.
De openbare verdediging zal plaatsvinden op
Woensdag 10 juni 2020 om 11:00 uur
door
Marjolein Yvonne Andrea Maria Kroonen
geboren op 30 oktober 1990
te Ermelo
Prof. dr. H.J. Lambers-Heerspink
Prof. dr. D. de Zeeuw
Copromotor
Dr. J. StevensBeoordelingscommissie
Prof. dr. R.O. DayProf. dr. P. Rossing Prof. dr. S.J.L. Bakker
Jeroen V. Koomen Nienke M.A. Idzerda
Chapter 1 Introduction and aims of this thesis 11
Chapter 2 A dosing algorithm for metformin based on
the relationships between exposure and renal clearance of metformin in patients with varying degrees of kidney function
25
Chapter 3 Exposure-response relationships of the
sodium-glucose co-transporter 2 inhibitor dapagliflozin for renal risk markers
47
Chapter 4 Population pharmacokinetics and individual
UACR exposure-response analysis for Empagliflozin, Linagliptin and Telmisartan in patients with type 2 diabetes
65
Chapter 5 Association between individual cholesterol and
proteinuria response and exposure to atorvastatin or rosuvastatin
91
Chapter 6 Summary and future perspectives 109
Chapter 7 Appendices
Nederlandse samenvatting en toekomstperspectieven Dankwoord
About the author
1
Introduction and aims
Published in adapted form
M.Y.A.M. Kroonen,
H.J. Lambers-Heerspink,
D. de Zeeuw;
In: (Clinical) Trial and Error
in Diabetic Nephropathy,
J.J.Roelofs, L.Vogt, (2019)
Status of trials in nephrology
Relatively low number of clinical trials in the field of nephrology
The landscape of clinical trials in general faces considerable challenges. Attrition rates in late stages of drug development are increasing along with a continuous rise in drug development costs. (1,2)
A concerted effort is necessary to find out how to develop new effective and safe drugs in an efficient and cost-effective way.
The area of nephrology, does not only suffer from the general problems faced in clinical trials, but also holds a number of other specific problems among which the smaller number of clinical trials in comparison to other specialties [Figure 1], most trials were too small to detect a realistic treatment effect, and sub-optimal quality of most trials. (3)
Figure 1. Number of published randomized controlled trials (RCT) in nephrology and twelve other specialties
of internal medicine from 1966 to 2002. (4)
Factors contributing to the low number of clinical trials in nephrology include the lack of visibility, lack of availability of new or more effective drugs, and the availability of patients willing to participate in clinical trials. (5-8) However, these low numbers do not adequately reflect the urgent need for new treatment strategies in nephrology.
Chronic kidney disease is a large public health problem
The number of people requiring dialysis for end-stage renal disease (ESRD) has been increasing rapidly across the world. (9) This increase closely parallels ongoing growth in the prevalence of diabetes for which it is estimated that the number of diabetic patients will increase further from 415 million in 2015 to 642 million by 2040. (10) Chronic kidney disease (CKD), in particular when advanced stages are reached, is associated with a high risk of premature mortality, has a huge impact on the quality of life of patients and their relatives, and places a heavy burden on national health care budgets. (11,12) The high numbers of affected patients would suggest a high awareness to develop and test new interventions and thus a high number of clinical trials. Instead the opposite is seen. However, new interventions are highly desired, particularly since the current guideline recommended strategy of targeting the renin– angiotensin–aldosterone system (RAAS), is of proven benefit in preventing and treating diabetic nephropathy for some patients, but by far not for all. (13,14)
Successful trials
Trials in the past decades of nephrology research have provided insights in the targets to treat patients with diabetes mellitus type 2 and nephropathy. Some trials have shown that tight glycemic control has delayed the onset and progression of nephropathy in patients with diabetes mellitus type 1 and 2. Blood pressure lowering, in particular with ACE-inhibitors or Angiotensin Receptor Blockers, have also been shown to delay the onset of ESRD in patients with type 2 diabetes and CKD. (15,16) Analyses from trials with ARBs have also provided more insight in the importance of albuminuria lowering as an additional target for treatment. (17,18)
Residual risk
Despite the promising and successful results from optimal RAAS inhibition in
combination with tight glycemic and blood pressure control, many patients with diabetes and nephropathy still progress to ESRD. (13) The high residual risk is illustrated by the fact that the reduction of end-stage renal disease in the RENAAL was only 28% and not 100%. (19) In itself, this 28% is a considerable risk reduction compared to conventional therapy. Yet, the starting absolute risk in this population was substantial and thus a high residual risk remains despite the large risk reduction. (Figure 2)
New drugs and new targets
The identified high residual risk requires new targets and therapies. Several options were tested: Further lower the known risk markers by increasing the inhibition of the RAAS with existing and with new drugs; further lower the known risk markers by new interventions targeting novel target and pathophysiological pathways.
Figure 2. Renal risk reduction in two renal outcome trials RENAAL (losartan) and IDNT (irbesartan) in
comparison with treated cancer (1996-2003; US Cancer Institute Surveillance Epidemiology and End Results database). The Y-axis indicates death with the assumption that in the absence of dialysis or renal
transplantation, patients with end-stage renal disease would die. RAS = renin-angiotensin-system. Summarizes the residual renal risks after interventions in the IDNT and RENAAL trial. (20). These data highlight the need for additional therapies that further lower the risk of ESRD. Ongoing research has focused on the discovery of new drugs and targets that lower renal risk markers on top of established treatments.
Increasing RAAS-inhibition
RAAS inhibition with mono-therapy has been shown to be renoprotective in patients with diabetes and nephropathy. The next step was to inhibit the RAAS more stringently by combining existing therapies in an effort to further lower blood pressure and albuminuria. Unfortunately, the trials demonstrated no additional benefit of more stringent RAAS blockade either with combination of ACE-inhibitors and ARBs or combination of direct renin inhibition as adjunct to ACE-inhibition or ARBs. In fact, combination therapy was associated with a higher risk of hyperkalemia and acute kidney
injury in all trials. Accordingly, current guidelines discourage the use of combination RAAS therapy.
New drugs same targets
New treatments to further lower albuminuria on top of single RAAS blockade were also discovered in the last decade. Sulodexide is one of these new treatments. Several studies showed that this drug lowered albuminuria on top of ACE-ARB by an allegedly effect on the glycocalyx, (a thin gel like layer covering the endothelium). (21-25) However, two large trials failed to demonstrate benefits on a surrogate outcome (albuminuria reduction) ESRD. The endothelin system is another promising target that has been targeted. Avosentan and atrasentan are examples of drugs that antagonize endothelin receptors. These agents marked promising results in their potential to lower blood pressure and albuminuria on top of RAAS inhibition in phase 2 trials. (26,27) These promising results led to large outcome trials such as the ASCEND, a trial in which the endothelin antagonist avosentan was studied in a clinical trial randomizing 1392 subjects to receive 25mg or 50mg avosentan or placebo. The ASCEND trial was terminated early because of safety issues particularly congestive heart failure due to the sodium retaining effects of endothelin antagonists. (28) However, the important lesson from the ASCEND trial was that a very specific endothelin receptor antagonist should be used to minimize the sodium retaining effects and mitigation strategies (such us diuretic use and selection of patients not prone to fluid retention) should be implemented. These strategies were implemented in the SONAR trial. The trial showed indeed that the endothelin receptor antagonist atrasentan showed a marked renoprotective effect with markedly lower incidence of heart failure compared to the ASCEND trial.(29)
New targets
Inflammation and oxidative stress emerged as important pathophysiological pathways that accelerate the development and progression of diabetic nephropathy. This knowledge has led to the development of specific anti-inflammatory anti-oxidative modulator such as bardoxolone methyl. (30) A phase 2 study with this agent demonstrated that bardoxolone causes a sustained increase in eGFR.
These promising results led to the design of a large outcome trial, designed to confirm the efficacy of bardoxolone methyl in 2185 patients with type 2 DM and CKD stage 4.
However, this study had to be terminated early again because of increased rates of congestive heart failure and mortality in the bardoxolone methyl arm.
Reason for these failed trials
The above summary is disappointing. None of the trials mentioned above have been able to identify a single new and effective treatment strategy for patients with diabetes and nephropathy despite the enormous human and financial resources that have been put into these large trials. Several important trial design elements may explain why the trials did not lead to the desired outcomes. It is unlikely that the trial endpoint and trial population were inappropriate since the endpoints and population were similar as the successful RENAAL and IDNT trials in the early 2000s. However, post-hoc analysis of these trials showed that in almost all trials lack of effect on the good risk markers (those leading to a good outcome) and/or too much effect on bad risk markers (those leading to poor outcomes) played an important role in the failure of these trials as summarized in Table 1.
For example, in a post hoc analysis of the ALTITUDE trial, it was shown that although aliskiren decreased blood pressure and albuminuria, there were still many patients in the aliskiren treatment arm who did not have a reduction in albuminuria. Moreover, patients with a robust lowering in albuminuria (i.e. >30% reduction) during the first six months of treatment had a significantly lower risk compared to patients in whom albuminuria did not change. (37) This suggests that if only these albuminuria responder patients were selected, the outcome of the trial would have been highly positive.
A similar situation was seen in the BEACON trial which was terminated after 9 months because of high rates of congestive heart failure and mortality. (36) However, exclusion of patients at high risk of congestive heart failure by selecting a population with a low BNP level (<200 pg/mL) and without a history of congestive heart failure gave a completely different picture. In this selected population, bardoxolone methyl actually did not increase the risk of congestive heart failure and may even offer renoprotection. (38) Selecting the right patient who thus tolerates the drug and beneficially respond to the drug is a key design element to conduct more efficient trials.
Ta ble 1. O ve rv ie w o f fa ile d t ria ls Tr ia l T ype St ud y a rm s P ri m ar y o ut co m e Re sul t F ai lur e O NT A RGE T (3 1) (n = 2 5 6 20 ) DM 1& 2 T el mis ar ta n ve rs us Ra m ip ril ve rs us C omb in at io n Dia ly si s, DS C R, de at h. T he n umb er o f ev en ts fo r t he c omp os ite p rima ry ou tc om e w as s im ila r fo r t el m is ar ta n ( n= 11 47 [1 3· 4%] ) a nd r am ip ril (1 15 0 [ 13 ·5 %] ; h aza rd ra tio [H R] 1 ·0 0, 9 5% C I 0 ·9 2– 1· 09 ), b ut w as in cr ea se d w ith c omb in at io n t he ra py (1 23 3 [1 4. 5%] ; H R 1 ·0 9, 1 ·0 1– 1· 18 , p = 0· 03 7) . E ffe ct o n su rr oga te b ut hi gh S ide e ff ec t SU N -M AC RO (3 2) (n = 1 24 8) DM 2 Su lo de xi de ve rs us Pl ac eb o bo th o n t op o f RAA S b lo ck ade D SC R, E SRD , o r se ru m c re at in in e 6 .0 m g/dl. T he s ul ode xi de gr ou p h ad a lo w er n um be r of pr ima ry e ndp oi nt s. B ut c omp ar is on w as n ot st at is tic al ly s ign ific an t ( ha za rd r at io : 0 .8 5 [ 95 % co nfide nc e i nt er va l: 0 .5 0– 1. 44] ; P = 0.54 ). N o e ff ec t o f su rr oga te T RE AT (3 3) (n = 4 03 8) DM 2 Da rb ep oe tin -α ve rs us Pl ac eb o E SRD , de at h o r a ca rdi ov as cu la r ev en t D ea th o r a c ar di ov as cu la r e ve nt (h aza rd r at io f or da rb ep oe tin a lfa v s. p la ce bo , 1 .0 5; 9 5% co nfide nc e i nt er va l [ C I] , 0 .9 4 t o 1 .1 7; P = 0 .4 1) D ea th o r e nd -s ta ge r en al di se as e i n da rb ep oe tin al fa v s p la ce bo g ro up (h az ar d r at io , 1 .0 6; 9 5% C I, 0 .9 5 t o 1 .1 9; P = 0 .2 9) . E ff ec t o n su rr oga te b ut hi gh s ide e ff ec t
ALT IT U D E (3 4) (n = 8 56 1) DM 2 Al is ki re n Ve rs us Pl ac eb o bo th o n t op o f RAA S bl oc ka de E SRD , D SC R, de at h o r t im e t o ca rdi ov as cu la r de at h / fir st oc cu rr en ce o f ca rdi ac a rr es t Aft er a m edi an fo llo w -u p o f 3 2. 9 m on th s, th e pr im ar y e nd p oi nt h ad o cc ur re d i n 7 83 p at ie nt s (1 8. 3% ) a ss ign ed to a lis ki re n a s c om pa re d w ith 73 2 ( 17 .1 % ) a ss ign ed to p la ce bo (h aza rd r at io , 1. 08 ; 9 5% c on fide nc e i nt er va l [ C I] , 0 .9 8 t o 1 .2 0; P = 0 .1 2) . Lo w e ff ec t o n su rr oga te b ut hi gh s ide e ff ec t VA N E P H RO N D (3 5) (n = 1 44 8) DM 2 Lis in op ri l ve rs us Pl ac eb o B ot h o n t op o f lo sa rt an E SRD , de at h o r t he fir st o cc ur re nc e o f a c ha nge in th e es tim at ed G F R, o r. C om bi na tio n t he ra py o ff er ed n o r en al b en efit bu t r es ul te d i n e xc es si ve ris k o f h yp er ka le m ia (6 .3 v er su s 2 .6 e ve nt s p er 10 0 p er so n y ea rs ; P < 0 .0 01 ) a nd a cu te k idn ey in ju ry (12.2 ve rs us 6. 7 e ve nt s p er 1 00 p er so n y ea rs ; P < 0 .0 01 ). Lo w e ff ec t o n su rr oga te a nd hi gh s ide e ff ec t B E AC O N (3 6) (n = 2 18 5) DM 2 B ar do xo lo ne m et hy l v er su s Pl ac eb o E SRD , o r de at h fr om c ar di ov as cu la r ca us es . pr im ar y c om po si te o ut co m e ( ha za rd r at io in th e ba rdo xo lo ne m et hy l gr ou p v s. th e p la ce bo gr ou p, 0 .9 8; 9 5% co nfide nc e i nt er va l [ C I] , 0 .7 0 to 1 .3 7; P = 0 .9 2) De at h fr om c ar di ov as cu la r c au se s o cc ur re d i n 2 7 pa tie nt s r an do m ly a ss ign ed t o b ar do xo lo ne m et hy l a nd in 1 9 r an do m ly a s ign ed t o p la ce bo (h aza rd r at io , 1 .4 4; 9 5% C I, 0 .8 0 t o 2 .5 9; P = 0.23) E ff ec t o n su rr oga te a nd hi gh s ide e ff ec t AS C E N D (2 8) (n = 1 39 2) DM 2 Av os en ta n 25 m g/5 0 m g ve rs us Pl ac eb o Al l gr ou ps o n to p of RAAS bl oc ka de D SC R, E SRD , o r de at h. Av os en ta n r edu ce d p ro te in ur ia c om pa re d w ith pl ac eb o, b ut , h ad e xc es s adv er se c ar di ov as cu la r ev en ts ; e sp ec ia lly flu id o ve rlo ad ( 4. 6%; P = 0.225 ), co nge st iv e he ar t fa ilu re (3 .6 %; P = 0 .1 94 ) an d de at h ( 2. 6%) . E ff ec t o n su rr oga te b ut hi gh s ide e ff ec t
Variability in drug response
Although the above examples are post-hoc analyses and are prone to bias and confounding, they illustrate that the design of the clinical outcome trials in diabetic nephropathy should pay more attention to the individual response of patients to existing and new drugs. In fact, response variation may play a much bigger and more important role than we thought in clinical trial design and appropriate attention to it may be a game changer when it comes to the design of new trials (i.e. personalized medicine).
Personalized medicine has been embraced in the oncology trials. Targeted therapies and careful patient selection based on genetic or biomarkers is common practice as illustrated by numerous examples. (39-41)
The nephrology area should follow the example from the oncology area and start selecting (and excluding) patients for trial enrollment who either do not tolerate the drug or who do not respond to the drug. The aforementioned SONAR trial is a first step in this direction. The design of the trial was such that patients are selected for trial participation based on their response to the drug. In this trial all patients received the endothelin receptor antagonist atrasentan for six weeks. Patients in whom albuminuria decreases by more than 30% (responders) and in whom there are no signs of sodium retention (e.g. no increase in body weight) were randomized to treatment with atrasentan or placebo. The trial demonstrated a marked renoprotection and a small increase in incidence of heart failure. Ongoing analyses from this trial will evaluate the usefulness and utility of the enrichment design both in terms of efficacy and safety.
A better understanding of the factors involved in the variability in response is needed to implement personalized medicine in clinical trials as well as in clinical practice. Drug response variation can be caused by variations in the genetic background of an individual that affect the drug target. For example, genetic variations in the ACE gene may affect disease progression and response to ACE-inhibitors and ARBs. Indeed, in clinical trials with these agents it has been shown that the ACE polymorphism DD genotype is associated with a faster disease progression and a more favorable response to ACE inhibitors (enalapril or ramipril) and the ARB losartan. (42,43)
Response variation may also be caused by between individual variations in systemic exposure to a drug which may be the consequence of underlying variations in absorption, distribution, metabolism or excretion. Previous studies focus on the relationship between drug dose and drug exposure with pharmacodynamic responses on a population level. Such studies investigate the optimal dose with optimal efficacy and minimal side effects on a population level but do not investigate the balance between efficacy and safety in individual patients and to what extent this varies among patients. There is thus a paucity of data in patients with diabetes and chronic kidney disease on the relationship between individual drug exposure and individual treatment response.
A better understanding of the between individual variability in drug exposure and pharmacodynamics response is needed followed by exploration of factors that determine the individual exposure. This will pave the way for specific interventions and strategies to optimize individual drug response with the ultimate aim to more efficiently design clinical trials and use drugs in clinical practice.
Research aims of this thesis
The important lessons learned from a decade of clinical trial failures in nephrology is that the one size fits all approach does not fit everyone. Accordingly, much more emphasis should be placed on the individual and how the individual responses to the prescribed drugs. As described above, various studies have investigated which clinical characteristics are involved. However, there are only few studies that systematically investigated to what extent individual drug exposure determines individual drug response in patients with type 2 diabetes. Therefore, the studies conducted in this thesis were designed to better understand the individual variation in drug response by investigating the role of individual drug exposure to various drugs registered for clinical use in patients with diabetes.
In Chapter 2 we aimed to investigate how the renal clearance (CLr) and apparent non-renal clearance of metformin (CLnr/F) in patients with varying degrees of kidney function influenced metformin exposure. This was performed in order to develop a dosing algorithm to tailor recommendations for patients using metformin.
In Chapter 3 we aimed to explore the exposure-response relationship for atorvastatin and rosuvastatin for LDL and UPCR. The PLANET trials randomized patient with a urine-protein to creatinine ratio (UPCR) of 500-5000 mg/g and a fasting LDL-cholesterol >2.33 mmol/L to a 52-week treatment period with atorvastatin 80 mg, rosuvastatin 10 mg or rosuvastatin 40 mg and showed that atorvastatin but not rosuvastatin was able to reduce UPCR at similar LDL lowering properties. The individual changes in both UPCR and LDL-cholesterol during treatments with these statins varied widely between patients. This inter-individual variability could not be explained by patients’ physical or biochemical characteristic. Therefore, we aim to assess whether the plasma concentrations of both statins are associated with LDL-cholesterol and UPCR response.
In Chapter 4 we aimed to investigate the exposure-response relationship for the sodium glucose co-transporter 2 inhibitor dapagliflozin for renal risk markers. Dapagliflozin has been shown to decrease various renal risk markers such as HbA1c, systolic blood pressure, body weight and albuminuria in patients with diabetes mellitus type 2. Prior studies have confirmed that the response in these renal risk markers is variable between patients but reproducible upon re-exposure. Suggesting the response to dapagliflozin is a true pharmacological response rather than a random response variation. We aimed to explain the response variability in the different renal risk markers to the exposure of dapagliflozin.
In Chapter 5 we aimed to explore the exposure-response relationship in albuminuria for empagliflozin, linagliptin and telmisartan. To evaluate this exposure response relationship where we define exposure as (AUC0-∞) we sampled according to an additional
pharmacokinetic protocol in the ROTATE-2 trial. In this trial, patients with diabetes mellitus type 2 and elevated albuminuria were randomized to 4 different albuminuria lowering treatments.
References
1. burden of progressive chronic kidney disease among patients with type 2 diabetes. J Diabetes Complications 2014 Jan-Feb;28(1):10-16.
2. Manns B, Hemmelgarn B, Tonelli M, Au F, So H, Weaver R, et al. The Cost of Care for People With Chronic Kidney Disease. Can J Kidney Health Dis 2019 Apr 4;6:2054358119835521.
3. Zoungas S, Chalmers J, Neal B, Billot L, Li Q, Hirakawa Y, et al. Follow-up of blood-pressure lowering and glucose control in type 2 diabetes. N Engl J Med 2014 Oct 9;371(15):1392-1406.
4. Strippoli GF, Craig JC, Schena FP. The number, quality, and coverage of randomized controlled trials in nephrology. J Am Soc Nephrol 2004 Feb;15(2):411-419.
5. Makino H, Haneda M, Babazono T, Moriya T, Ito S, Iwamoto Y, et al. The telmisartan renoprotective study from incipient nephropathy to overt nephropathy--rationale, study design, treatment plan and baseline characteristics of the incipient to overt: angiotensin II receptor blocker, telmisartan, Investigation on Type 2 Diabetic Nephropathy (INNOVATION) Study. J Int Med Res 2005 Nov-Dec;33(6):677-686.
6. Hellemons ME, Persson F, Bakker SJ, Rossing P, Parving HH, De Zeeuw D, et al. initial angiotensin receptor blockade-induced decrease in albuminuria is associated with long-term renal outcome in type 2 diabetic patients with microalbuminuria: a post hoc analysis of the IRMA-2 trial. Diabetes Care 2011 Sep;34(9):2078-2083.
7. Schievink B, de Zeeuw D, Parving HH, Rossing P, Lambers Heerspink HJ. The renal protective effect of angiotensin receptor blockers depends on intra-individual response variation in multiple risk markers. Br J Clin Pharmacol 2015 Oct;80(4):678-686.
8. de Vries JK, Levin A, Loud F, Adler A, Mayer G, Pena MJ. Implementing personalized medicine in diabetic kidney disease: Stakeholders' perspectives. Diabetes Obes Metab 2018 Oct;20 Suppl 3:24-29. 9. Heerspink HJ, Kropelin TF, Hoekman J, de Zeeuw D, Reducing Albuminuria as Surrogate Endpoint (REASSURE) Consortium. Drug-Induced Reduction in Albuminuria Is Associated with Subsequent Renoprotection: A Meta-Analysis. J Am Soc Nephrol 2015 Aug;26(8):2055-2064.
10. Petrykiv SI, Laverman GD, de Zeeuw D, Heerspink HJL. The albuminuria-lowering response to dapagliflozin is variable and reproducible among individual patients. Diabetes Obes Metab 2017 Mar 14.
2
A dosing algorithm for metformin
based on the relationships between exposure
and renal clearance of metformin in patients with
varying degrees of kidney function
J.K. Duong M.Y.A.M. Kroonen S.S. Kumar H.J. Lambers-Heerspink C.M. Kirkpatrick G.G. Graham K.M. Williams R.O. Day
Published in European Journal of Clinical Pharmacology.
2017 Aug; 73(8):981-990
Abstract
Purpose: The aims of this study were to investigate the relationship between metformin
exposure, renal clearance (CLR), and apparent non-renal clearance of metformin
(CLNR/F) in patients with varying degrees of kidney function and to develop dosing
recommendations.
Methods: Plasma and urine samples were collected from three studies consisting of
patients with varying degrees of kidney function (creatinine clearance, CLCR; range, 14
mL/min – 112 mL/min). A population pharmacokinetic model was built (NONMEM) in which the oral availability (F) was fixed to 0.55 with an estimated inter-individual
variability (IIV). Simulations were performed to estimate the AUC0-τ, CLR, and CLNR/F.
Results: The data (66 patients, 327 observations) were best described by a
two-compartment model and CLCR was a covariate for CLR. Mean CLR was 17 L/h (CV 22%)
and mean CLNR/F was 1.6 L/h (69%). The median recovery of metformin in urine was
49% (range 19 – 75%) over a dosage interval. When CLR increased due to improved
renal function, AUC0-τ decreased proportionally, while CLNR/F did not change with
kidney function. Target doses (mg/day) of metformin can be reached using CLCR/3x100
to obtain median AUC0-12 of 18 – 26 mg/h/L for metformin IR and AUC0-24 of 38 - 51
mg/L/h for metformin XR, with Cmax <5 mg/L.
Conclusions: The proposed dosing algorithm can be used to dose metformin in patients with various degrees of kidney function to maintain consistent drug exposure. However, there is still marked IIV and therapeutic drug monitoring of metformin plasma
Introduction
Metformin is the first-line pharmacotherapy in the treatment of type 2 diabetes mellitus (T2DM). Metformin has an excellent safety profile, with favourable properties including weight neutrality and no increased risk of hypoglycaemia. The long-term use of
metformin is also associated with a reduction in the risk of diabetes-related deaths (1) and the risk of some cancers. (2)
Metformin is cleared by the kidneys, and a dose-proportional reduction with renal function is recommended to reduce the risk of adverse effects such as lactic acidosis. (3) The accumulation of metformin has been previously assessed using peak
concentrations of metformin (Cmax <5 mg/L) (4,5), and we have proposed dosage
regimens of metformin at various stages of renal function to maintain Cmax <5 mg/L. (6)
The pharmacokinetic parameters of immediate release (IR) metformin include a moderate and variable oral availability (55 + 16% mean ± SD) in healthy subjects. (7) The fractional availability of the extended release (XR) metformin is very similar. (8) There is, however, large inter-individual variability (IIV) in the estimate of the total
clearance of metformin (CLTOTAL/F) which is influenced by its oral availability (F). The
renal clearance of metformin (CLR), on the other hand, is not affected by F and,
therefore, is expected to have a lower IIV than CL/F.
In the present study, we have utilized population pharmacokinetic approaches
to investigate the proportion of metformin cleared by the kidneys (CLR), the proportion
of the drug not cleared by the kidneys (non-renal clearance of metformin, CLNR/F) and
the drug exposure (AUC0-12,AUC0-24) of metformin in a large sample of patients with
varying degrees of renal function.
Methods Patients
Patients receiving metformin (immediate-release, IR; extended-release, XR) for T2DM were recruited from the outpatient Diabetes Clinic at St Vincent’s Hospital (Australia) and Ziekenhuis Groep Twente Hospital (Almelo, The Netherlands). Demographic characteristics and data on medical comorbidities and concurrent medications were collected. Informed consent was obtained from all individual participants included in the study. Studies 1 and 2 were approved by the Human Research Ethics Committee at St Vincent’s Hospital and University of New South Wales, Sydney (08209/SVH08/035;
09280/SVH09/080), and were registered with the Australian New Zealand Clinical Trials Registry (ACTRN12611000908932).
study 3 was approved by the Ethics Committee of the University Medical Center Groningen (Groningen, The Netherlands; METc 2013.178).
Study design
This was an observational, open-label, cross-sectional study consisting of intensive and sparse blood and urine sampling designs from three studies (study 1 – 3, Figure 1). Blood samples were collected to determine metformin and lactate concentrations in plasma, as well as serum creatinine and HbA1c concentrations. Urine samples were collected to determine metformin and creatinine concentrations. All patients were taking metformin in the long term for the treatment of T2DM.
Study 1
Study 1 was an intensive blood sampling study of patients with a range of kidney function attending the Diabetes Clinic, St Vincent’s Hospital (Sydney, Australia). Information on dosing regimen, dosage and times of last dose prior to blood sampling was collected. All patients were admitted to an observation ward for the study. Patients treated with metformin IR (n = 7) provided blood samples at 0 (pre-dose), 0.5, 1, 2, 3, 4, 6, 8, 10 and 12 h after metformin administration. Urine samples were collected from 0 to 12 h after a regular dose. The blood sampling times for patients treated with metformin XR (n = 9) were 0, 2, 4, 6, 12, 16, 20 and 24 h after their metformin dose. Urine samples for patients treated with metformin XR were collected from 0 to 24 h post dose.
Study 2
Study 2 was a sparse blood sampling study design of patients with chronic kidney disease
(CKD; creatinine clearance, CLCR <40 mL/min) attending the Diabetes Clinic at St
Vincent’s Hospital (Sydney, Australia). These patients (n = 5) participated in an
interventional study of metformin conducted over 6 weeks (4) and were prescribed daily doses of 500 mg metformin IR. Information on dosing regimen, dosage and times of last dose was collected. A total of eight blood samples and timed urine samples (2 – 4 h) were collected from each patient. The patients noted the time of their last void, and the 2-hour urine sample was collected at the clinic.
Study 3
Study 3 was conducted in patients with mild to moderate kidney disease (n = 45, CLCR <
60 mL/min) attending an outpatient Diabetes Clinic at Ziekenhuis Groep Twente Hospital (Almelo, The Netherlands).
Prior to entering the study, patients withheld their usual dose of metformin. The baseline sample was then collected for metformin determination (pre-dose), and patients voided their bladder. After taking their usual dose of metformin, a 2-hour blood sample was collected to determine the concentration of metformin in plasma. Urine was collected over 24 hours and metformin and creatinine concentrations measured. Patients did not take another dose of metformin during this time interval. The patients collected urine samples at home.
Information on the patient’s dosing regimen was collected; however, the times of last dose were not recorded. Therefore, some assumptions were made when dealing with these data (see “Missing dosage times” section).
Figure 1. Study design of study 1, study 2 and study 3. The black dots are the blood sampling time points and
the grey shaded area is the urine collection interval.
Metformin assay
Metformin concentrations in plasma and urine samples were analysed at the Department of Clinical Pharmacology and Toxicology, St Vincent’s Hospital, Australia. For samples collected at Almelo Hospital, plasma and urine samples were stored at -4°C before transport to St Vincent’s Hospital. The plasma concentrations of metformin were assayed by HPLC using a validated method. (6) Urinary concentrations were assayed similarly.
Population modelling
The plasma concentration data and the accumulated amount of metformin in the urine (mg) were used for population pharmacokinetic analyses. The data were analysed by non-linear mixed effect modelling using NONMEM® version 7.2 (ICON Development Solutions, Ellicott City, MA, USA) with the first-order conditional estimation method with interaction (FOCE-I). For nested models, model selection was informed by using the objective function value (OFV, -2log likelihood), whereby a decrease of >3.84 was
considered statistically significant (P <0.05, χ2 distribution). The Akaike information
criterion (AIC) was used to select the best model between non-nested models. Model runs were executed using Perl-Speaks-NONMEM 3.5.3 (9), model development was managed using Pirana 2.8.1 (10) and all plots were generated with R (Version 3.2.0). (11)
Missing dosage times
Because the time of the last dose of metformin was not recorded in study 3, some assumptions regarding the dosing history were necessary for the population
pharmacokinetic analyses. Firstly, all subjects using metformin were assumed to be at steady-state, receiving metformin IR and compliant with their metformin dosing. Secondly, dosage intervals were assumed to be equal, i.e., doses taken every 24 h for once daily dosing, every 12 h for twice daily dosing and every 8 h for thrice daily dosing. Lastly, the patients were asked to withhold their scheduled dose of metformin prior to the start of the study. This was assumed to be the dose of metformin immediately prior to the start of the study. These assumptions were evaluated by investigating the
agreement between the observations and the predictions, with and without using the pre-dose time-point for study 3.
Structural and statistical model
A previously developed population PK model was used. (6) This was a
two-compartment model with first-order absorption (ka) for metformin IR, and zero-order
absorption for metformin XR (D) (Figure 2). The renal clearance of metformin (CLR)
was estimated using the accumulated amount of metformin excreted in the urine. Unabsorbed drug together with clearance by all other means is termed the non-renal
Figure 2. The final two-compartment model of metformin describing first-order absorption for metformin
immediate-release (IR) and zero-order absorption for metformin extended release (XR), where ka is the
first-order absorption constant and D is the duration of infusion. CLR, is the renal clearance, CLNR/F is the
non-renal clearance, Q/F is the inter-compartmental clearance, VC/F is the apparent central volume of distribution,
and VP/F is the apparent peripheral volume of distribution.
The following parameters were estimated by the model: CLNR/F, CLR, central volume of
distribution (VC/F), inter-compartmental clearance (Q/F) and peripheral volume of
distribution (VP/F). The total clearance of metformin (CLTOTAL/F) was calculated using
CLR and CLNR/F. From previous studies, the median value of F was 55%. (7,12)
Therefore F was fixed to a value of 0.55 with an estimated IIV. The IIV for the pharmacokinetic parameters was described using a log-normal distribution:
𝑃𝑖= 𝑃𝑇𝑉 × 𝑒𝑥𝑝(𝑛𝑖) (1)
where Pi is the pharmacokinetic parameter of the ith individual, PTV is the typical
population parameter value, ni describes the variability between the ith individual and the
population parameter which follows a normal distribution with a mean of 0 and a
variance of ω2 i.e. N(0, ω2).
Several residual error models were tested for metformin concentrations in the plasma and urine. Different error models were tested for each study to account for inaccuracies in sample collection.
The combined additive and proportional residual models (mixed error models) were described as:
𝐶𝑖𝑗 = 𝐶(1 + 𝜀𝑝𝑖𝑗) + 𝜀𝑎𝑖𝑗 (2)
where Cij is the jth measured observation for the ith individual and εpij and εaij are the
proportional and additive residual random errors, respectively, for the ith individual and
jth measurement. εpij had a normal distribution of N(0, σ12) and εaij had a normal
distribution of N(0, σ22), where σ1 and σ2 represent the standard deviations for the
proportional and additive residual error, respectively.
A time-dependent error model (13) was also tested for study 3 due to the missing dosage history. This was investigated as a step function for the plasma concentrations, whereby a different error model was tested for the pre-dose samples, compared to the 2-h post-dose samples.
Covariate model
Covariates for CLR, CLNR/F and VC/F were screened by inspecting scatter plots of
empirical Bayes estimates (EBEs) against characteristics of patients (age, sex, weight, lean body weight (14)) and between studies. The equation used for calculating lean body weight was as follows (14):
𝐿𝐵𝑊(𝑚𝑎𝑙𝑒) = 9.27 x 103 x WT
6.68 x 103 x 216 x BMI (3)
𝐿𝐵𝑊(𝑓𝑒𝑚𝑎𝑙𝑒) = 9.27 x 103 x WT
8.78 x 103 x 244 x BMI (4)
where WT is total body weight (kg), and BMI is the body mass index (kg/m2).
Creatinine clearance was tested as a covariate for all clearance parameters and was
calculated using the Cockcroft-Gault equation (CLCR) (15) with either total body weight,
Creatinine clearance was also calculated directly from the urine output/plasma
concentration (UV/P,Eq. 5, CLUCR) and was tested as a covariate for CLR.
𝐶𝐿𝑈𝐶𝑅=
𝐶𝑟𝑒𝑎𝑡𝑖𝑛𝑖𝑛𝑒𝑢𝑟𝑖𝑛𝑒 (𝜇𝑚𝑜𝑙𝐿 ) 𝐶𝑟𝑒𝑎𝑡𝑖𝑛𝑖𝑛𝑒𝑠𝑒𝑟𝑢𝑚(𝜇𝑚𝑜𝑙𝐿 )
× 𝑉𝑜𝑙𝑢𝑚𝑒𝑢𝑟𝑖𝑛𝑒 (𝑚𝐿)
𝑇𝑖𝑚𝑒(ℎ𝑜𝑢𝑟𝑠)× 60 (5)
Stepwise covariate modelling was utilized to select significant covariates and included the forward selection (OFV <3.84 points, P <0.05, d.f. = 1) and backward elimination of potential covariates (OFV <6.63 points, P <0.01, d.f. = 1).
Model evaluation
The models were evaluated by inspecting diagnostic plots and prediction-corrected visual predictive checks (VPCs, n = 1000). The VPC of the final model was evaluated by
comparing the 10th, 50th and 90th percentiles of the observations to the corresponding
10th, 50th and 90th percentiles of the simulations (n = 1000). (17) A non-parametric
bootstrap (n = 1000) stratified by study was used to evaluate the uncertainty of all pharmacokinetic parameter estimates in the final model to obtain the 95% confidence interval for all parameters, as described previously. (18)
Dosing simulations
The final model with residual variability and parameter uncertainty was used to simulate
1000 concentration-time profiles at steady-state doses of 500 mg IR and at CLCR of 30
mL/min, 60 mL/min, 90 mL/min and 120 mL/min. The model-derived AUC0-τ was
used to investigate the relationship between drug exposure and CLCR. Relationships
between CLTOTAL/F, CLR and CLCR were also investigated.
Results
Patients
A total of 66 patients with T2DM were investigated (Table 1). In study 1, data collection stopped prematurely for two patients taking metformin XR once daily. For these two patients, serial plasma and urine samples were only collected up until 12 h and 8 h post-dose, respectively. There were no significant differences in the HbA1c values and plasma lactate concentrations between the three studies (Table 1).
Pharmacokinetics
A total of 327 observations from the 66 patients were used for population pharmacokinetic analyses. Over a dosage interval (study 1), the median recovery of metformin in the urine was 49% (19 – 75%, range) of the dose. The dataset was best described using a two-compartment model with first-order absorption for IR and
zero-order absorption for XR. Inter-individual variability (IIV) was added on to the CLNR/F,
CLR, VC/F and F parameters, and covariance of CLNR/F and CLR were accounted for in
the model. The inclusion of IIV for F reduced the IIV of CLNR/F from 172% to 69%.
A mixed error model was used to describe the residual error for the plasma concentrations of metformin. The model was further improved by using different error models for the urine output for subjects from study 3 (study 1 and study 2 vs study 3, ΔOFV -24.2). A mixed residual error model best described the residual error for the urine concentrations for studies 1 and 2, while an additive residual error best described the residual error for study 3. A time-dependent error model was investigated for Study 3, but it did not improve the model predictions.
Despite the lack of dosing information for study 3, there was good agreement between observed and predicted concentrations (without the inclusion of the pre-dose data points) and the eta and epsilon shrinkage was low (<30%), suggesting that the assumptions regarding the times of last doses were valid.
Table 1. Patient characteristics by study.
Parameter Study 1 Study 2 Study 3 Total
N 16 5 45 66
Age (years) 64 (40–79) 73 (64–80) 69 (51–79) 68 (40–80)
Weight (kg) 84 (67–165) 80 (60–127) 99 (64–156) 99 (60–165)
Body mass index
(kg/m2) 29 (25–45) 31 (23–42) 33 (23–49) 32 (23–49)
Lean body weight (kg) 60 (43–87) 60 (44–72) 65 (41–80) 63 (41–87)
HbA1c (%) 7.0 (5.5–14.5) 7.1 (5.8–10.1) 6.5 (5.1–8.6) 7.0 (5.5–14.5)
Lactate (mmol/L) 1.7 (1.0–2.9) 0.9 (0.8–3.0) 1.7 (0.6–3.7) 1.7 (1.0–2.9)
Creatinine clearance
(mL/min) 75 (24–112) 26 (14–38) 44 (28–75) 46 (15–112)
The CLCR using lean body weight was found to be the most significant covariate for CLR and reduced IIV from 45% to 22% (P <0.01, Eq. 6):
𝐶𝐿𝑐𝑟(𝐿/h) = (𝑆𝑒𝑟𝑢𝑚 𝑐𝑟𝑒𝑎𝑡𝑖𝑛𝑖𝑛𝑒 (𝜇𝑚𝑜𝑙 𝐿140−𝐴𝑔𝑒(𝑦𝑒𝑎𝑟𝑠)× 𝐿𝐵𝑊(𝑘𝑔)⁄ )× 0.814) × 0.06 (6)
CLCR estimated using Cockcroft-Gault and with lean body weight resulted in a lower
OFV than CLUCR. Lean body weight reduced the IIV for VC/F by 4% but it was not
significant at P <0.01 following backwards elimination of the covariate. Other body size descriptors (total body weight, lean body weight) were not significant covariates for any of the pharmacokinetic parameters.
The final model equation for CLR is as follows (Eq. 7):
𝐶𝐿𝑅= 𝜃𝐶𝐿𝑅 × (
𝐶𝐿𝐶𝑅
2.7) × 𝑒
𝜂𝐶𝐿𝑅 (7)
where 𝜃𝐶𝐿𝑅 is the mean population value for CLR, CLCR was centered to the median
population value of 2.7 L/h (45 mL/min) and ηCLR is the IIV for CLR.
The final model showed good agreement between the observations and the model predictions (Figure 3) and good agreement between the observations and the model simulations (Figure 4). Due to the limited number of urine samples collected in the 0- to 12-h post-dose period, wide 95% confidence intervals were observed for the accumulated urine output (Figure 4). All population pharmacokinetic parameter estimates and their precision are summarized in Table 2. A large residual error was observed for the urine concentrations from study 3, which may be due to the patients’ variable compliance to the study protocol.
At metformin doses of 500 mg IR, there was a proportional increase in drug exposure
with reduced renal function (Table 3). The CLR of metformin increased approximately in
proportion to CLCR. The median ratio of CLR to CLCR was 6.6 (5.1 – 8.8, 5-95%), and
median ratio of CLTOTAL/F to CLCR was 13 (9 - 36). The CLNR/F was much smaller than
Figure 3. Diagnostic plots of the final model. The plots of observations against population predictions and
individual predictions, were plotted with the line of identity (black) and a linear regression line (blue). Plots of conditional weighted residuals (CWRES) were plotted with a locally weighted scatterplot smoothing curve (red).
Figure 4. Visual predictive checks (VPCs) of metformin concentrations in plasma (top) and cumulative
amount excreted in urine (bottom). The lines represent the 10th, 50th and 90th percentiles of the observations
and the shaded areas represent 95% confidence intervals of the simulated concentrations of the 10th, 50th and
Table 2. The population parameter estimates and the median parameter values (5– 95%) of the non-parametric bootstrap replicates of the final pharmacokinetic model.
Parameter Estimate (RSE %) Median (5–95%) Structural parameters CLNR/F (L/h) 1.6 (65) 1.2 (0.1–3.1) CLR (L/h) 17 (7) 17 (16–19) VC/F (L) 123 (22) 118 (71–179) VP/F (L) 335 (42) 336 (180–1038) Q/F (L/h) 13 (29) 12 (7.3–19) k a (1/h) 0.51 (30) 0.47 (0.28–1.1) D (h) 9.6 (8) 9.6 (7.4–14.4) F 0.55 (FIX) –
Inter-individual variability (IIV)
CLNR/F (CV%)a 69 (49) 83 (25–206)
CLR (CV%) 22 (21) 22 (14–30)
VC/F (CV%) 36 (23) 36 (18–54)
F (CV%) 29 (16) 28 (18–36) Residual error model
Proportional error, plasma (CV%) 20 (43) 20 (13–26) Additive error, plasma, SD (mg) 0.05 (118) 0.05 (0.001–0.09) Proportional error, urine,
studies 1 and 2 (CV%) 6.7 (93) 6.4 (0.0007–11.9) Additive error, urine, studies 1 and 2, SD (mg) 24 (70) 23 (8–36) Additive error, urine, study 3, SD (mg) 112 (35) 109 (71–140)
RSE is the percentage calculated as standard error divided by mean estimate
CLNR/F non-renal clearance, CLR renal clearance, VC/F central volume of distribution, VP/F peripheral volume of distribution, Q/F apparent inter-compartmental clearance, ka first-order absorption, D duration of infusion, F bioavailability
Target doses of metformin (while maintaining median Cmax <5 mg/L, (6)) can be reached using the dosing algorithm (Eq.8):
𝐷𝑜𝑠𝑒 (𝑑𝑎𝑦𝑚𝑔) = 𝐶𝐿𝐶𝑅
3 × 100 (8)
This dosing algorithm was used to simulate concentration-time profiles by CLCR (Figure
5). Since the maximum recommended XR dose of metformin is 2000 mg, this dose was
used for dosing simulations at CLCR of 60 to 90 mL/min. Using this dosing algorithm,
median AUC0-12 for metformin ranged from 18 to 26 mg/L/h (Table 3). Similarly, the
median AUC0-24 for metformin XR ranged from 38 to 51 mg/L/h. Therefore, this
dosing algorithm provided a consistent drug exposure across a range of renal function as
well as maintaining the median Cmax to be below 5 mg/L (Table 3).
Figure 5. Simulated median plasma concentrations of metformin for CLCR of 15 mL/min to 30 mL/min, 60 to
90 mL/min, and 90 to 120 mL/min (IR only) at daily doses of 500 mg, 1000 mg, 2000 mg (maximum XR dose), and 3000 mg, respectively.
T ab le 3 . D os in g s imu la tio ns o f m et fo rm in IR ( tw ic e d ail y) an d m et fo rm in X R ( on ce d ail y) d os es at v ar io us s ta ge s of k idn ey fu nc tio n. 15 to 3 0 mL /mi n 30 to 6 0 mL /mi n 60 to 9 0 m L /m in 90 to 1 20 mL /mi n IR D os e ( m g/1 2 h) 250 500 1000 1500 AU C 0– 12 (m g/L /h ) 19 (1 1– 30 ) 20 (1 4– 31 ) 26 (1 7– 38 ) 18 (1 2– 28 ) C ma x (m g/L ) 1. 4 ( 0. 6– 2. 1) 1. 5 ( 0. 9– 2. 7) 2. 3 ( 1. 2– 4. 3) 1. 8 ( 1. 0– 3. 4) CL R (L /h ) 9 ( 6– 13 ) 17 (1 2– 24 ) 29 (2 0– 41 ) 40 (2 8– 58 ) CL NR /F ( L /h ) 1. 6 ( 0. 5– 4. 8) 1. 6 ( 0. 5– 4. 8) 1.5 (0 .5 –4. 6) 1. 6 ( 0. 5– 4. 8) XR D os e ( m g/da y) 500 1000 2000 a AU C 0-24 (m g/L /h ) 38 (2 2– 59 ) 41 (2 7– 63 ) 51 (3 4– 73 ) C ma x (m g/L ) 1. 5 ( 0. 8– 2. 7) 1. 9 ( 1. 1– 3. 5) 2. 7 ( 1. 5– 4. 9) CL R (L /h ) 9 ( 6– 13 ) 17 (12 –25 ) 29 (21 –42 ) CL NR /F ( L /h ) 1. 6 (0 .5 –5. 3) 1. 5 ( 0. 5– 4. 8) 1. 6 ( 0. 5– 5. 0) M edi an (5 -95 % ) a T w o t ho us an d m ill ig ram s i s t he m ax im um r ec om m en de d dai ly do se f or X R. D os ag e w ith I R tab le ts is r ec om m en de d i f hi gh er do se s a re r eq ui re d.
Discussion
It is well accepted that metformin doses should be reduced in patients with impaired renal function because the major mode of elimination is urinary excretion of the unchanged drug.
This study provides further support for a dose-proportional reduction in metformin doses with the decline in kidney function. Based on the recommendation to keep
metformin Cmax below 5 mg/L (5,6), we have proposed a dosing algorithm to estimate
appropriate doses by renal function. The maximum recommended daily dose of metformin is 3000 mg for IR and 2000 mg for XR. (19) In our simulations, daily doses
of 2000 mg XR were used for CLCR of 60 to 90 mL/min (XR only). If higher doses are
required, patients should be switched from metformin XR to metformin IR, provided that they do not experience gastrointestinal side-effects. (19) For metformin IR, median
AUC0-12 was 18 to 26 mg/L/h, which is similar to AUC0-24 for metformin XR (38 to 51
mg/L/h) for patients with CLCR from 15 to 90 mL/min. However, the ranges of peak
plasma concentrations and AUC values are very wide (Table 3) and we therefore suggest that the plasma concentrations should be measured in order to optimize dosage with this important drug.
This is the first population pharmacokinetic model of metformin to describe
the relationship between CLR, CLNR/F and drug exposure in patients with T2DM and
kidney disease and CLCR estimated using Cockcroft-Gault and with lean body weight was
a significant covariate for CLR. Unlike our previous model (6), body weight was not a
covariate for VC/F. This is likely due to the high shrinkage for VC/F (37%), which may
have hidden the true covariate relationship with weight. (20) The median value of F in our study was 49 with (range, 19 to 75%). This result was similar to the values of 55 ± 19% (mean ± SD). (12)
There are limited and conflicting results on the non-renal clearance of metformin. Previous studies on the urinary recovery of metformin after the reference intravenous injection have reported either complete recovery (21) or 80% recovery in the urine, with no metformin recovered in the faeces (12). By comparison, our
pharmacokinetic analysis indicates that the CLNR/F is much smaller than CLR. Further
studies are required to investigate whether a non-renal elimination pathway exists for metformin in man. In rats, the metabolism of metformin was suggested due to the
reduced half-life of metformin by cytochrome P450 enzymes and lengthened by
inhibitors of these enzymes but no corresponding study has been conducted in man. (22)
The ratio of CLR and CLTOTAL/F with CLCR were higher than estimated
previously. In our review of the pharmacokinetics of metformin, the ratio of CLR to
CLCR was 4.0 ± 1.5 (7) compared to a median value of 6.6 (5 to 95% range 5.1 to 8.8) in
the present studies. Further, the ratio of the CLTOTAL/F in our review was 10.7 ± 3.5
which is considerably lower than the ratio, 13.1 (5 to 95% range 9.3 to to 36.5) in the present studies. The reasons for the contrast are unclear, but many patients included in our previous review either did not have T2DM or were administered single doses of metformin. By contrast, all patients in the present study were dosed with metformin in
the long term for their T2DM. Furthermore, the wide range of CLTOTAL/CLCR in the
present study is probably due to the large inter-patient variation of F.
There were some limitations to the present study. The shrinkage on VC/F was
high due to the sparse collection of pharmacokinetic time-points for Studies 2 and 3,
which may have masked the true covariate effect of weight for VC/F. Additionally, there
were a limited number of urine samples collected over the dosage interval and all urine collections were assumed to be complete. study 1 was conducted entirely in the research centre while patients enrolled in study 2 were required to report the times of last void and patients from study 3 collected their 24-h urine at home. Poor compliance to the study protocol would contribute to the variability in the estimate of urinary recovery.
Potential compliance issues for study 3 were accounted for using a separate residual error model for the urine concentrations. This error model revealed a large additive residual error for study 3, which significantly improved the model and indicated poor compliance to urine collection at home.
Conclusion
We have described the CLR, CLNR and AUC of metformin in patients with varying
degrees of renal function. We have proposed a dosing algorithm that can be used to
reduce metformin doses proportionally with kidney function to maintain Cmax below 5
mg/L and to achieve a consistent drug across a range of renal function. CLR was well
estimated, however further studies are required to obtain reliable estimates of CLNR/F
References
1. Holman RR, Paul SK, Bethel MA, Matthews DR, Neil HA (2008) 10-year follow-up of intensive glucose control in type 2 diabetes. N Engl J Med 359 (15):1577-1589.
2. Chen YC, Kok VC, Chien CH, Horng JT, Tsai JJ (2015) Cancer risk in patients aged 30 years and above with type 2 diabetes receiving antidiabetic monotherapy: a cohort study using metformin as the comparator. Ther Clin Risk Manag 11:1315-1323.
3. Duong JK, Furlong TJ, Roberts DM, Graham GG, Greenfield JR, Williams KM, Day RO (2013) The Role of Metformin in Metformin-Associated Lactic Acidosis (MALA): Case Series and Formulation of a Model of Pathogenesis. Drug Saf. 36 (9), 733-46
4. Duong JK, Roberts DM, Furlong TJ, Kumar SS, Greenfield JR, Kirkpatrick CM, Graham GG, Williams KM, Day RO (2012) Metformin therapy in patients with chronic kidney disease. Diabetes, obesity & metabolism.
5. Lalau JD, Lemaire-Hurtel AS, Lacroix C (2011) Establishment of a database of metformin plasma concentrations and erythrocyte levels in normal and emergency situations. Clinical drug investigation 31 (6):435-438.
6. Duong JK, Kumar SS, Kirkpatrick CM, Greenup LC, Arora M, Lee TC, Timmins P, Graham GG, Furlong TJ, Greenfield JR, Williams KM, Day RO (2013) Population pharmacokinetics of metformin in healthy subjects and patients with type 2 diabetes mellitus: simulation of doses according to renal function. Clinical pharmacokinetics 52 (5):373-384.
7. Graham GG, Punt J, Arora M, Day RO, Doogue MP, Duong JK, Furlong TJ, Greenfield JR, Greenup LC, Kirkpatrick CM, Ray JE, Timmins P, Williams KM (2011) Clinical pharmacokinetics of metformin. Clinical pharmacokinetics 50 (2):81-98.
8. Timmins P, Donahue S, Meeker J, Marathe P (2005) Steady-state pharmacokinetics of a novel extended-release metformin formulation. Clinical pharmacokinetics 44 (7):721-729.
9. Lindbom L, Pihlgren P, Jonsson EN (2005) PsN-Toolkit--a collection of computer intensive statistical methods for non-linear mixed effect modeling using NONMEM. Comput Methods Programs Biomed 79 (3):241-257.
10. Keizer RJ, van Benten M, Beijnen JH, Schellens JH, Huitema AD (2011) Pirana and PCluster: a modeling environment and cluster infrastructure for NONMEM. Comput Methods Programs Biomed 101 (1):72-79.
11. R Core Team (2015). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL http://www.R-project.org/. R Core Team (2015). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL http://www.R-project.org/.
12. Tucker GT, Casey C, Phillips PJ, Connor H, Ward JD, Woods HF (1981) Metformin kinetics in healthy subjects and in patients with diabetes mellitus. Br J Clin Pharmacol 12 (2):235-246
13. Karlsson MO, Beal SL, Sheiner LB (1995) Three new residual error models for population PK/PD analyses. Journal of pharmacokinetics and biopharmaceutics 23 (6):651-672
14. Janmahasatian S, Duffull SB, Ash S, Ward LC, Byrne NM, Green B (2005) Quantification of lean bodyweight. Clinical pharmacokinetics 44 (10):1051-1065.
15. Cockcroft DW, Gault MH (1976) Prediction of creatinine clearance from serum creatinine. Nephron 16 (1):31-41
16. Janmahasatian S, Duffull SB, Chagnac A, Kirkpatrick CM, Green B (2008) Lean body mass normalizes the effect of obesity on renal function. Br J Clin Pharmacol 65 (6):964-965.
17. Karlsson MO, Holford N. A tutorial on visual predictive checks. PAGE 2008, Abstr 1434 [www.page-meeting.org/?abstract=1434].
18. Henderson AR (2005) The bootstrap: a technique for data-driven statistics. Using computer-intensive analyses to explore experimental data. Clin Chim Acta 359 (1-2):1-26.
19. Australian Medicines Handbook. Adelaide: Australian Medicines Handbook Pty Ltd; 2017. Goswami S, Yee SW, Stocker S, Mosley JD, Kubo M, Castro R, Mefford JA, Wen C, Liang X, Witte J, Brett C, Maeda S, Simpson MD, Hedderson MM, Davis RL, Roden DM, Giacomini KM, Savic RM (2014) Genetic variants in transcription factors are associated with the pharmacokinetics and
20. Savic RM, Karlsson MO (2009) Importance of shrinkage in empirical bayes estimates for diagnostics: problems and solutions. The AAPS journal 11 (3):558-569.
21. Pentikainen PJ, Neuvonen PJ, Penttila A (1979) Pharmacokinetics of metformin after intravenous and oral administration to man. Eur J Clin Pharmacol 16 (3):195-202
22. Choi YH, Lee MG (2006) Effects of enzyme inducers and inhibitors on the pharmacokinetics of metformin in rats: involvement of CYP2C11, 2D1 and 3A1/2 for the metabolism of metformin. British journal of pharmacology 149 (4):424-430.
3
Exposure-response relationships of the
sodium-glucose co-transporter 2 inhibitor
dapagliflozin for renal risk markers
M.Y.A.M. Kroonen J.V. Koomen S.I. Petrykiv G.D. Laverman H.J. Lambers-Heerspink J. Stevens
Published in Diabetes Obesity and Metabolism
2020 Jan 27 [Epub ahead of print]
Abstract
Introduction: Dapagliflozin is a sodium-glucose co-transporter 2 inhibitor which is developed as oral glucose lowering drug but has also been shown to decrease other renal risk markers including blood pressure and albuminuria. Prior studies have demonstrated that the individual response in these renal risk markers varies between patients and is reproducible upon re-exposure. In this study, we quantitated the consistency in
individual plasma exposure to dapagliflozin upon re-exposure, and investigated whether the individual systemic exposure to dapagliflozin explained the between individual variation in response to dapagliflozin in multiple renal risk markers.
Methods: Data were used from a cross-over randomized clinical trial that assessed the albuminuria lowering effect of dapagliflozin in 33 patients with type 2 diabetes and elevated albuminuria. Fifteen patients were exposed twice to dapagliflozin. Trough plasma concentrations of dapagliflozin were measured for each patient at steady state. Dapagliflozin plasma concentrations were measured by liquid chromatography tandem mass spectrometry (LC-MS/MS) and pharmacokinetic parameters were simulated based on a population pharmacokinetic model. Linear-mixed effects models were used to quantify the exposure-response relationships.
Results: Median plasma concentration after first and second exposure to dapagliflozin was 5.3 ng/mL vs 4.6 ng/mL respectively (p=0.78). Lin’s concordance correlation coefficient between occasions was 0.73 (p <0.0021). Every 100 ng.hr/mL increment in area under the dapagliflozin plasma concentration curve was associated with a decrease in log-transformed urinary albumin:creatinine ratio 5.9, p<0.01), body weight (β=-0.3, p<0.01), eGFR (β=-0.7, p=0.01) and an increase in UGE (β=17.0, p<0.001). Conclusion: An individuals exposure to dapagliflozin is consistent upon re-exposure and correlates with pharmacodynamic response in renal risk markers.
Introduction
Sodium-glucose co-transporter 2 (SGLT2) inhibitors are a relatively new class of oral glucose lowering drugs that have been approved for the treatment of type 2 diabetes mellitus. Dapagliflozin is a SGLT2 inhibitor, which has been shown to lower glycated haemoglobin (HbA1c) by promoting urinary glucose excretion. (1) In addition, dapagliflozin also decreases body weight, systolic blood pressure and albuminuria. A large cardiovascular outcome trial demonstrated that dapagliflozin also significantly reduced the risks of heart failure and progression of chronic kidney disease in patients with type 2 diabetes. (2,3)
Despite the observed beneficial effects on a population level, individual patients show a large inter-individual variation in response to SGLT2 inhibitors, as assessed by varying degrees of changes in renal risk markers. (4,5) This inter-individual response variation is reproducible upon re-exposure, suggesting that the individual variation in drug response is not a random variation in the surrogate marker but a real
pharmacological response variation. (4) Previous studies have not identified clinical variables that could explain all inter-individual response variability, neither baseline characteristics nor genetic polymorphisms in the SGLT2 gene. (4,6)
In addition to inter-individual variation in pharmacodynamic variables, a study pooling pharmacokinetic data from 20 studies in both healthy volunteers and patients with type 2 diabetes mellitus also showed inter-individual variation in pharmacokinetic parameters to SGLT2 inhibitors. (7) In addition, previous studies have associated the systemic exposure of SGLT2 inhibitors with changes in urinary glucose excretion, as these responses directly reflect the mechanism of action. (7-9) However, there is limited information on the exposure-response relationship between systemic exposure of SGLT2 inhibitors and changes in other renal risk markers such as systolic blood pressure and albuminuria. To investigate this relationship, it is important firstly to demonstrate the consistency in dapagliflozin exposure after re-exposing the same patient in order to ensure that the exposure is a true pharmacological response and not random.
The aim of the present study, therefore was, to quantitate consistency in plasma exposure to dapagliflozin upon re-exposure. Subsequently, we investigated whether the individual systemic exposure to dapagliflozin explained the inter-individual variation in response to dapagliflozin with regard to multiple renal risk markers.
Methods
Clinical trial design and patient population
Data were used from the IMPROVE trial, a prospective, randomized, double-blind, placebo-controlled, cross-over clinical trial that evaluated the albuminuria-lowering effect of dapagliflozin and the reproducibility of this effect within patients. The study design, patient population and main results have been published previously. (10) In short, 33 patients with type 2 diabetes, an urine albumin to creatinine ratio (UACR) between 100 mg/g and 3500 mg/g (11.3-395,5 mg/mmol), an estimated glomerular filtration rate (eGFR) ≥ 45 mL/min/1.73m2 and aged between 18 and 75 years were enrolled. Patients were required to be treated with a maximum tolerated dose of an angiotensin converting enzyme inhibitor or angiotensin receptor blocker for at least 4 weeks. Patients were randomized to three 6-week treatment periods; dapagliflozin 10 mg/day, placebo, dapagliflozin 10 mg/day or vice versa with a wash-out period of 6 weeks between treatment periods. The study was registered with the Dutch Trial Register (NTR 4439) and was performed in accordance with the Declaration of Helsinki.
Measurements
Patients collected 3 consecutive first morning void urine samples on day -2, day -1 and day 0 for measurement of urinary albumin and creatinine at baseline and at every first and last visit per treatment period. UACR data were calculated as the geometric mean from the 3 first morning void urine collections. Body weight and systolic blood pressure (SBP) were registered at every visit. eGFR was calculated using the Modification of Diet in Renal Disease equation. (11) Urinary glucose excretion was measured in 24-hour urine samples. Samples for dapagliflozin plasma concentration were taken per protocol at 24
hours after first dosing but before the second dose (Ctrough). Plasma concentrations of
dapagliflozin were measured for all patients (n=33) by a previously validated Liquid Chromatography Mass Spectroscopy technique at Covance laboratories (Indianapolis, IN, USA). Of these, two samples were excluded based on concentration below the lower
limit of quantification (n=2). Patients with a measured Ctrough plasma concentration that
exceeded the 90% confidence interval (CI) of simulated Ctrough were excluded from
analysis because they had most likely taken their study medication prior to sample collection.
