How to initiate lithium therapy: a systematic review of dose estimation and level prediction methods

Journal of Affective Disorders 146 (2013) 15–33

Contents lists available at SciVerse ScienceDirect

Journal of Affective Disorders journal homepage: www.elsevier.com/locate/jad

Review

How to initiate lithium therapy: a systematic review of dose estimation and level prediction methods Sienaert P.a,b,n, Geeraerts I.a, Wyckaert S.a a

Department of Mood Disorders, University Psychiatric Center, Catholic University Leuven, Campus Kortenberg, Leuvensesteenweg 517, 3070 Kortenberg, Belgium b ECT Department, University Psychiatric Center, Catholic University Leuven, Campus Kortenberg, Leuvensesteenweg 517, 3070 Kortenberg, Belgium

a r t i c l e i n f o

abstract

Article history: Received 17 May 2012 Received in revised form 9 August 2012 Accepted 10 August 2012 Available online 1 September 2012

Background: Throughout the past decades, several methods have been developed to achieve therapeutic lithium blood levels as quick and safe as possible. The present study will systematically review the methods developed and studied for lithium dose estimation or level prediction at the initiation of therapy. Methods: A systematic computerized Medline search was performed for papers published in English, French or Dutch between 1966 and April 2012 describing or studying methods for dosing lithium or predicting the lithium level on a certain dosage. References of relevant articles were screened for additional papers. Results: Of 273 unique references retrieved, 65 met the inclusion criteria. Apart from the empirical titration method, 38 predictive methods for initiating lithium were identified. These methods can be classified into two categories: the a priori predictive methods, and the test-dose predictive methods requiring the administration of a test dose of lithium prior to starting treatment. Limitations: The methodological strength was not taken into account for a study to be included in the review. Conclusions: The most important distinction between the empirical titration method and the predictive methods appears to be the shorter time the latter need to achieve the targeted lithium level. The vast majority of predictive methods, however, show inconsistent or poor results or have not been replicated since their initial description. The empirical titration method, although not extensively studied, appears to be a time-honored method that can be recommended for use in daily clinical practice. & 2012 Elsevier B.V. All rights reserved.

Keywords: Lithium Therapeutic drug monitoring Bipolar disorder Predictive dosing methods Empirical dosing Systematic review

Contents 1. 2. 3. 4.

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 4.1. Methodological concerns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 4.2. Practical applicability. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 4.3. Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 5. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 Role of funding source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 Conflict of interest. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

1. Introduction n

Corresponding author at: University Psychiatric Center, Catholic University Leuven, Campus Kortenberg, Leuvensesteenweg 517, 3070 Kortenberg, Belgium. Tel.: þ32 2 7580511; fax: þ32 2 7595380. E-mail address: [email protected] (P. Sienaert). 0165-0327/$ - see front matter & 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.jad.2012.08.013

Since the psychiatric use of lithium was first described in 1949 (Cade, 1949), lithium has become an established therapeutic agent for the treatment of both acute manic, mixed and depressive

16

P. Sienaert et al. / Journal of Affective Disorders 146 (2013) 15–33

episodes and the maintenance treatment of bipolar disorder (APA, 2002). In addition, it is often used as an augmentation agent in antidepressant-resistant major depressive disorder and in prophylaxis of recurrent depressive disorders, schizoaffective psychosis and pathologic aggressive behavior (Bauer et al., 2006). Despite wide experience and proven efficacy, the use of lithium has not remained undisputed, mainly because of potential renal impairment and dangerous toxicity. The specific pharmacokinetic and pharmacodynamic properties of lithium complicate its therapeutic use (Malhi et al., 2011, 2012). Lithium has a narrow therapeutic index and there is a wide interindividual variation in renal clearance of lithium and response to treatment (Grandjean and Aubry, 2009). This implies that for each patient, an appropriate dose has to be determined, and regular monitoring of blood levels has to be ensured. The most widely used method to start lithium therapy is the clinical titration method (Bauer et al., 2006). Lithium is started at a low dose, and a 12-hour serum level is measured after steady-state concentrations are reached, i.e. after one week. The daily dose is then adjusted, with gradual increments, to reach the desired serum level. This procedure is based on the linear relationship between lithium dose and blood levels at steady-state (Bauer et al., 2006).

Throughout the past decades, several methods have adressed the question of how to achieve therapeutic blood levels as quick and safe as possible. These methods provide formulas to predict the dosage an individual patient will require to achieve a preset lithium blood level or calculate the expected steady state level on a certain dosage of lithium. Seventeen of these predictive methods were reviewed before, in 1988 (Lobeck, 1988). It was concluded that all methods had considerable shortcomings and should be used with due caution (Lobeck, 1988). The present study will systematically review all methods for lithium dose estimation or level prediction at the initiation of therapy, developed and/or studied to date. The clinical usefulness of these lithium initiation methods, i.e. their accuracy and practical value, will be discussed.

2. Methods A systematic computerized Medline search was performed for papers published between 1966 and April 2012, using the search terms ‘lithium’ and ‘dose/dosage prediction’, ‘dose/dosage

Table 1 Characteristics of a priori predictive methods. Method

Analytical approach

Parameters included

Number of articles studying the method

Pepin et al. (1980)

M/PK D

12

Sampath et al. (1981) Zetin et al. (1983)

LRA (187) LRA (100)

Lesar et al.( 1985)

LRA (71)

Kook Loading dose (Kook et al., 1985) Zetin et al. (1986) Higuchi Masonna (Higuchi et al., 1988) Groves et al. (1991)

NM

Creatinine clearance (Cockcroft–Gault, with use of ideal body weight) Body weight Lithium formulation (carbonate, citrate, extended release), use of TCA, age, sex, body weight Sex, weight, age, depression/use of TCA, state (acute/nonacute), creatinine clearance (Cockcroft-Gault) Sex, weight, age, depression/use of TCA, state (acute/nonacute) Body weight

LRA (548) M/PK D

Age, body weight, inpatient/outpatient, sex, use of TCA Body weight

18 1

Based on Stokes’ 76 observationsb NONMEM (79)

Body weight

2

Lean Body weight (calculated from body weight and height), creatinine clearance (Cockcroft-Gault) Clinical considerations (age, body size, past history of tolerance,y)

4

Age, body weight, serum creatinine level

2

Time since last dose, serum creatinine level

l

Age, body weight, BUN Body weight

4 2

Body weight, creatinine clearance (Cockcroft-Gault) Age, body weight, sex, BUN, use of TCA, creatinine clearance (Cockcroft-Gault), inpatient/outpatient Age, body weight, sex, serum creatinine level

1 2

Jermain NONMEM (Jermain et al., 1991) Moscovich Loading dose (Moscovich et al., 1992) Yukawa NONMEM (Yukawa et al., 1993) Sproule Fuzzy logic (Sproule et al., 1997) Terao et al., 1999 Keck Loading dose (Keck et al., 2001) Chiu et al. (2007) Abou-Auda et al. (2008) Huang et al. (2008)

NM NONMEM (90 patients, 303 data sets) Fuzzy logic modeling (10 patients, 87 data sets) LRA (70) M/PK D M/PK D LRA (60) M/PK D

4 6 1 1 1

1

1

NM: not mentioned in the article. LRA: linear regression analysis based upon patient data (the number of patients used for analysis is noted between brackets). M/PK D: mathematical/pharmacokinetic derivation. NONMEM: non-linear regression analysis with the Non-linear Mixed Effects Model Program, based upon patient data (the number of patients used for analysis is noted between brackets). Fuzzy logic modeling: predictions based on the creation of a knowledge base of prediction rules by the principles of fuzzy logic modeling, (the number of patients used for creation of the knowledge base is noted between brackets). Loading dose: methods used to give a loading dose of lithium, to rapidly attain therapeutic levels, in the case of acute mania. BUN: blood urea nitrogen. TCA: tricyclic antidepressants. a Mean lithium population pharmacokinetic parameters by Mason et al. were used to derive estimates of lithium clearance and volume of distribution, dependent on the body weight of the patient. b Manic patients were designated to alternatingly placebo, low, medium or high doses of lithium chloride, as calculated according to body weight. High and medium dosages appeared to be more efficacious in improving manic ratings.

P. Sienaert et al. / Journal of Affective Disorders 146 (2013) 15–33

predicting’, ‘dose/dosage estimation’, ‘level prediction/predicting/ estimation’, ‘dose/dosage titration’, ‘empirical dosing’, ‘dosing model/method’, and ‘dosage requirements’. The titles and abstracts of the papers thus retrieved were read to determine their relevance to the review. If regarded relevant, the full-text paper was screened to meet the following inclusion criteria: (1) the paper describes a method for dosing lithium or predicting the lithium level on a certain dosage, or the paper studies one or more of these methods; (2) the methods described are used to initiate lithium therapy (methods used to predict lithium levels in patients on a stable treatment are not included); (3) lithium is used as a mood stabilizer; (4) the paper is written in English, French or Dutch and (5) studies adult human subjects. Data published as posters or book chapters are not included. References of all relevant articles were then screened for additional papers.

3. Results The search strategy yielded a total number of 273 unique references (lithium AND dose/dosage prediction (42/46), dose/ dosage predicting (35/36), dose/dosage estimation (45/40), level prediction/predicting/estimation (40/21/15), dose/dosage titration (23/16), empirical dosing (3), dosing model/method (15/70), dosage requirements (31)). Fifty-one manuscripts fulfilled the inclusion criteria. Cross-references yielded another fourteen papers meeting the inclusion criteria. For one (Poust et al., 1976) of the references, the full-text paper could not be retrieved. Apart from the empirical titration method, 38 different methods to initiate lithium are described. These methods were classified into two categories. The ‘a priori’ methods propose a model for initiating lithium, based upon patient characteristics, with or without information obtained after a blood sample, without administering a test dosage of lithium. The ‘test-dose’ methods base predictions upon the lithium level obtained after the administration of one or more test dosages. ‘A priori’ and ‘test-dose’ methods are summarized and described in Tables 1 and 2. Practically, the described methods encompass different formulations: equations (e.g. the methods of Zetin et al. (1983) and Abou-Auda et al. (2008)), nomograms (e.g. the methods of Cooper et al. (1973) and Rosenberg et al. (1987)), graphs (e.g. the method of Swartz (1991)) or computerized calculations (e.g. Bayesian (Taright et al., 1994), NONMEM (Yukawa et al., 1993) and Fuzzy logic (Sproule et al., 1997) methods) for predicting the lithium dosage or level. Tables 3 and 4 provide an overview of the studies testing the various methods. The results are grouped by method, not by study. Comparative results are included in the tables only when the cited article does not mention results for the specific methods separately. The tables indicate whether the study actually applied the method to initiate lithium therapy, or alternatively, compared calculations based on the formula with the actual levels/ doses obtained upon initiation with another technique. In both cases, both prospective and retrospective study designs are envisaged. The predictive performance of a method is best studied by mean prediction error and (root) mean square prediction error. Mean prediction error (the mean of the predicted minus observed values) is a measure of bias, while (root) mean square prediction error (the (root of the) mean of the squared difference between predicted and observed values), and the related mean absolute error, present measures of precision (Sheiner and Beal, 1981). However, these measures are scarcely reported by the studies (e.g. the mean prediction error for lithium level prediction in test-dose methods is reported in only 9 out of 38 data sets). These measures are therefore not included separately in the tables, but in the column ‘other values and results of interest’. In addition, the success rate, if reported, was included in a separate column.

17

4. Discussion Apart from the empirical titration method, 38 different methods to initiate lithium were retrieved by a systematic literature search. This high number is to some extent relative, however, since some of the methods are only minor modifications of previously published methods. A number of both methodological and practical concerns are of importance in comparing these various methods. 4.1. Methodological concerns A number of studies were flawed by methodological shortcomings. For the derivation of the Zetin method, patients were included which had not attained steady-state levels (Bryant et al., 1984). The very low and very high dosing categories in the Cooper nomogram appeared to be only based on a few patients’ data and were merely developed by extrapolation from the other data points (Browne et al., 1988). Marr et al. (1983) proposed a ‘repeated one-point’ method for use in patients where normal pharmacokinetic assumptions do not apply, but the authors tested their method in only six patients with none of them actually showing the altered pharmacokinetics for which this method was meant. Moreover, some of the methods may be criticized for making implausible predictions. This applies more specifically to the method of Zetin—by making abstraction of the law of conservation of matter (Swartz, 1984)—and the method of Sampath—a body weight method whose predicted dosages were paradoxically found to decrease again beyond a body weight of 80 kg (Valecha et al., 1990). Since lithium is excreted primarily through the kidneys, the lack of a measure of renal function, as in the methods of Zetin (Terao et al., 1995) and Keck (Lu et al., 2002), seems to be an important shortcoming. When kidney function is taken into account, however, some measures of renal function are better suited than others. Using BUN (blood urea nitrogen), as in the method of Terao, might account for the biased results as BUN seems to be much more influenced by extra-renal factors than serum creatinine (Chiu et al., 2007). Repeated one-point methods, e.g. the methods described by Marr et al. (1983) and Nelson (1988), or multiple-point methods, e.g. Perry’s method (Perry et al., 1982), on the other side, make it possible to calculate the elimination rate constant of lithium clearance for each specific patient. These methods may therefore be recommended for use in patients suspected with abnormal renal function or when normal pharmacokinetic assumptions probably do not apply, like with coadministration of other drugs or changes in volume of distribution or imbalance in electrolytes (Browne et al., 1988; Marr et al., 1983; Nelson, 1988). Another methodological shortcoming is the fact that studies generally have not taken into account the influence of diurnal variation and of posture and state on the glomerular filtration rate (GFR), and thus on the lithium clearance (Lauritsen et al., 1981; Perry et al., 1984). Perry et al. (1984) suggest that failures in their predictions might be due to variations in GFR (due to altered mental and physical activity state) between the moment the test dose was administered and the moment the steady-state concentration was determined. It may also account for the results of Williams et al. (1989), who found no difference between different sampling time points in a study of acutely manic patients with seriously disturbed sleep-wake patterns. Some of the methods calculate the lithium elimination rate constant, which cannot be done accurately over a time interval that is shorter than the lithium half-life (Nelson, 1988). Both factors play a role in the repeated one-point method of Nelson (1988) and the multiple-point method of Perry et al. (1982), who

18

P. Sienaert et al. / Journal of Affective Disorders 146 (2013) 15–33

Table 2 Characteristics of test-dose predictive methods. Predictive method

Analytical approach

Test dose (mg)

Number of test doses

Number of blood samples

Blood sampling time Other parameters included (hours after test dose)

Cooper et al. (1973) Gaillot et al. (1979)

LRA (425) M/PK D

600 NM

1 NM

1 NM

24 h NM

Slattery et al. (1980)

M/PK D

NM

1

1

Tyrer et al. (1981)

LRA (24)

1

1

Norman et al. (1982) Perry et al. (1982)

M/PK D M/PK D

1 1

1 2 or 3

10 h 12 and 36 h or 12, 24, 36 h

Marr Repeated one-point (Marr et al., 1983)

2 with 12 h dosing interval

2

Perry et al. (1983)

Principles of (Ritschel and Thompson (1979) M/PK D LRA (17)

1000 or 1200 (half the dosage if 455 years old) 600 600, 900, 1200 or 1500 (dependent on age and weight) 600

Within the range of half-lives encountered in de population 10 or 17 h

1200

1

Perry et al. (1984)

LRA (49)

1200

Perry et al. (1986) Karki Corrected multiple point of Perry ’82 (Karki et al., 1987)

LRA (17) modification of Perry ‘82: based on experimental observations M/PK D

Lobeck Modified Slattery (Lobeck et al., 1987) Rosenberg Modified Cooper (Rosenberg et al., 1987) Higuchi Bayesian (Higuchi et al., 1988)

Number of articles studying the method

None 12 Lithium clearance (determined by 2 urine measurements) Population elimination rate 3 constant 1

Lithium clearance (based on urine output during 7h) can be used instead of a blood level 4 h urine collection None

1 5

11 h after each test dose

None

1

1

24

1

1

24 h

1200 1200 (slow release formulation)

1 1

1 2 or 3

12, 24 or 36 h 12, 24 and 36 h

Indication treatment Indication treatment None None

900

1

1

24 h

modification 900 of Cooper: M/ PK D Bayesian (45) NM

1

1

24 h

NM

Z1

NM

NM

NM

Z1

NM

600

2

Z2

Iterative (45) Higuchi Fixed volume of distribution iterative method (Higuchi et al., 1988) Nelson Repeated oneM/PK D point (Nelson, 1988)

Williams Bayesian (Williams et al., 1989)

Bayesian (21) 1200

2 with 12 h or 24 h dosing interval 1

Williams Non-linear regression (Williams et al., 1989)

M/PK D

1200

1

Z3

Valecha et al. (1990) Swartz (1991)

LRA (60) M/PK D

900 NM

1 1–3 doses during 24 h

1 1 (or 2)

27 mmol

1

1

Taright Bayesian (Taright Bayesian et al., 1994) (113)

(prophylactic use/ of acute mania) (prophylactic use/ of acute mania)

Population elimination rate constant, height, age, weight, serum creatinine, sex None

2 1 5 1

2

2

1

Population values of lithium pharmacokinetic parameters (based on Mason ‘78 or Pepin ‘80) None

1

12 h after each test dose

None

1

Different time combinations possible, with 12 h intervals: i.e. 0, 12, 24, 36 h Different time combinations possible, with 12 h intervals: i.e. 0, 12, 24, 36 h 24 h Between 12 and 24 h (6 to 34 h if only 1 test dose is given)

Population values of lithium pharmacokinetic parameters

1

None

1

Body weight Lithium half-life: estimated according to age or derivated from 2 serum samples after a test dose Age, weight, height, serum creatinine level

1 1

24 h

1

NM: not mentioned in the article. LRA: linear regression analysis based upon patient data, (the number of patients used for analysis is noted between brackets). M/PK D: mathematical/pharmacokinetic derivation. Bayesian: use of a bayesian equation for estimation of parameters, (the number of patients used for analysis is noted between brackets). Iterative: use of an iterative computer program for estimation of parameters with a fixed volume of distribution and one concentration measurement after a test dose, (the number of patients used for analysis is noted between brackets). Non-linear regression: non-linear regression analysis, (with the algorithm REVOL) for estimation of pharmacokinetic parameters.

Table 3 Studies examining a priori predictive methods. Predictive method

Articles

Study design

Number of patients

Number of data sets

Outcome parameters

Results Success rate

Pepin et al. (1980)

Other values and conclusions of interest

Rate

Criterium of success or targeted goal

Calculation

71

71

D

64%

o 1 capsule/d deviation

(Patrias and Moore, 1985)

Calculation

23

23

D

11/23

(Lobeck et al., 1987)

Calculation

16

16

C

23/23 NM

*rounded predicted dose equal to actual dose * o 600 mg/d deviation NM

(Siemsen and Pasley, 1988)

Application

480

480

D/C

NM

NM

(Higuchi et al., 1988)

Calculation

45

219

C

NM

NM

(Browne et al., 1988)

Calculation

20

20

C

45%

o 0.2 mEq/l deviation

(Yukawa et al., 1993)

Calculation

34

60

C

NM

NM

NM

NM

D

(Wright and Crismon, 2000) Calculation

47

47

C D

(Stip et al., 2001)

Application

13

13

D/C

84%

0.6–1.2 mmol/l

(Chiu et al., 2007)

Calculation

129

129

D

NM

NM

(Abou-Auda et al., 2008)

Calculation

60

60

Cl D

NM

NM

*In 59% of patients an underprediction of 4 1 capsule/d occured and in 14% an overprediction of 41 capsule/d *In 10/71 patients a deviation of 43 capsules/d occured (in 9/10 patients this was an underprediction) *A significant underprediction of dosage occured in women NM

19

*MPE: 0.295 with CI: (0.045;0.545) *Mean squared error (mEq/l)2: 0.294 with CI ( 0.040;0.627) *Random variability of prediction occured *In obese patients with total body weight 4 30% greater than lean body weight, an underprediction of dosage occured *MPE: 0.147 with CI: (0.114;0.181) *RMSE: 0.320 with CI: (0.284;0.352) *Prediction errors were positively correlated with age and serum creatinine *Median prediction error : 0.095 with CI: (  0.14;0.27) *MAE: 0.22 *For concentration: MPE: 0.01 with CI: (  0.039;0.051) and MAE: 0.13 with CI: (0.106;0.162) *For dosage: MPE: 23.4 with CI: (  32.7;79.5) and MAE: 164.7 with CI: (128.5;200.9) *For concentration: MPE:  0.09 with CI: (  0.18;0.005) *For dosage: MPE:  187.69 with CI: (  299.78;-75.61) A significant underprediction of dosage occured and a significant difference between observed and predicted concentrations was observed *Toxicity group : MPE :  166.4 with CI: (  210.1;  122.7) *Control group: MPE:  126.0 with CI: (  189.9;  102.5) *For clearance in inpatients: MPE:  0.618 with SD: 0.103; MAE: 0.68 with SD: 0.10; RMSE: 0.95 with SD: 0.54; percent error:  24.36% with SD: 3.50 *For clearance in outpatients: MPE:  0.437 with SD: 0.071; MAE: 0.52 with

P. Sienaert et al. / Journal of Affective Disorders 146 (2013) 15–33

(Dugas and Feeney, 1983)

20

Table 3 (continued ) Predictive method

Articles

Study design

Number of patients

Number of data sets

Outcome parameters

Results Success rate Rate

Zetin et al. (1983)

Criterium of success or targeted goal SD: 0.055; RMSE: 0.624 with SD: 0.255; percent error: 21.39% with SD: 3.34 *For dosage in inpatients: MPE: 143.13 with SD: 74.39; MAE: 357.97 with SD: 57.37; RMSE: 534.92 with SD: 355.38; percent error: 13.39% with SD: 6.39 *For dosage in outpatients: MPE: 208.74 with SD: 82.93; MAE: 374.48 with SD: 66.79; RMSE: 546.5 with SD: 110.06; percent error: 18.47 % with SD: 7.20 *MPE: 0.27 and RMSE: 0.37 *Error of prediction: 36.57% *Sensitivity in detecting lithium levels outside therapeutic range: 80% *Specificity in detecting lithium levels outside therapeutic range: 76.19%

(Radhakrishnan et al., 2012) Calculation

31

31

C

NM

NM

(Gangadhar et al., 1989a)

Application

35

35

C

71%

0.8–1.2 mEq/l

(Gangadhar et al., 1989b)

Application

32

32

C

(Valecha et al., 1990)

Calculation

46

46

D

69% 72% NM

0.8–1.2 mEq/l at day 4 0.8–1.2 mEq/l at day 6–8 NM

(Srisurapanont et al., 2000)a Calculation

17

17

D

NM

NM

(Zetin et al., 1983)

100

100

D

112

112

71

71

66/100 94/100 64% 92% 68%

o 300 mg/d o 600 mg/d o 300 mg/d o 600 mg/d o 300 mg/d

93%

o 600 mg/d deviation

(Lesar et al., 1985)

Calculation

Calculation

D

deviation deviation deviation deviation deviation

(Patrias and Moore, 1985)

Calculation

23

23

D

6/23

(Lobeck et al., 1987)

Calculation

16

16

C

23/23 NM

*rounded predicted dose equal to actual dose * o 600mg/d deviation NM

(Browne et al., 1988)

Calculation

20

20

C

60%

o 0.2 mEq/l deviation

(Higuchi et al., 1988)

Calculation

45

219

C

NM

NM

*10% of patients had a subtherapeutic concentration *9% of patients had a concentration 41.8 mEq/l and signs of toxicity 3/32 patients showed clinical signs of toxicity *Negative correlation found between body weight and daily dose (as in the original findings of Sampath) *Theoretical analysis of predicted doses: a very narrow range of doses is suggested and the suggested doses decrease again beyond a body weight of 80 kg *MPE: 116.62 with CI: (6.63;226.62) and RMSE: 238.06 with CI: (150.83;300.99) *In 8 patients dosing errors 4149 mg/d (‘‘predictable error’’) occured MPE : 113 with SD: 307

*MAE: 256 with SD: 199 *2/71 patients had 4900 mg/d deviation *Predicted doses ranged from 45.6% less than to 45.9% greater than actual doses NM

*MPE:  0.055 with CI: (  0.163;0.054) * Mean squared error (mEq/l)2: 0.042 with CI (0.010;0.074) Median prediction error :  0.155 with CI: (  0.27;0.08)

P. Sienaert et al. / Journal of Affective Disorders 146 (2013) 15–33

Sampath et al. (1981)

Other values and conclusions of interest

MPE :  0.056 with CI: ( 0.098;  0.014) and RMSE: 0.322 with CI: (0.290;0.351) 71

71

D

75% 70/71

o 300 mg/d deviation o 600 mg/d deviation

(Lesar et al., 1985)

Calculation

71

71

D

69% 70/71

o 300 mg/d deviation o 600 mg/d deviation

Kook et al. (1985)

(Kook et al., 1985)

Application

38

38

C after 12 h

NM

NM

Zetin et al. (1986)

(Zetin et al., 1986)

Calculation

390

390

D

(Rosenberg et al., 1987)

Calculation

19

19

D

67.7% 92.8% 10/19

o 300 mg/d deviation o 600 mg/d deviation o 300 mg/d deviation

(Cummings et al., 1988)

Application

4

4

C

NM

NM

(Siemsen and Pasley, 1988)

Application

480

480

C

NM

NM

(Browne et al., 1989)

Calculation

20

20

C

50%

o 0.2 mEq/l deviation

(Zetin et al., 1990)

Calculation

104

104

D

62.5% 92.3%

o 300 mg/d deviation o 600 mg/d deviation

(Valecha et al., 1990)

Calculation

46

46

D

21/46

o 150 mg/d deviation

(Markoff and King, 1992)

Application

12

12

C

83%

o 0.1 mEq/l deviation

(Yukawa et al., 1993)

Calculation

34

60

C

NM

NM

29/29

0.5–1.3 mmol/l

D

(Cummings et al., 1993)

Application

29

29

C

*MAE: 233 with SD: 128 *Predicted doses ranged from 29.1% less than to 43.1% greater than actual doses *MAE: 252 with SD: 152 *Predicted doses ranged from 30.1% less than to 50.7% greater than actual doses *For males: MPE:  0.11 with SD: 0.03 and MAE: 0.16 with SD: 0.09 *For females: MPE:  0.04 with SD: 0.07 and MAE: 0.28 with SD: 0.14 *None of the patients showed clinical signs of toxicity *If obese females were left out: normal weight females had a MPE  0.15 (þ /  0.06) mEq/l. *Recalculation of concentrations for the obese females based upon ideal body weight gives a MPE of  0.23 (þ /  0.17) mEq/l MPE: 19 with SD: 325 *6/19 patients had a deviation between 300 and 600 mg/d *3/19 patients had a deviation of 4 600 mg/d Moderately accurate predictions with both over and underprediction The method showed a tendency to overdose, especially at the high end of the dosage range and in patients with compromised renal function Median prediction error : 0.066 with CI: (  0.09;0.16) *The method overestimated dose requirements for patients on lower doses and underestimated doses for patients requiring more than 1500 mg/d *MPE or MAE (mg/d): 174 with SD 271 *17/46 patients had a deviation of 150– 300 mg/d (13 overprediction, 4 underprediction) *8/46 patients had a deviation of 4 300 mg/d (3 overprediction, 5 underprediction) * only 2/12 patients achieved not the desired level, with an underprediction of the concentration occuring in these patietns *14 blood determinations were needed for the 12 patients *for concentration: MPE:  0.02 with CI: (  0.062;0.013) and MAE: 0.12 with CI: (0.093;0.139) *for dosage: MPE: 279.2 with CI: (221.8;376.6) and MAE: 302.0 with CI: (253.1;350.8) *MPE:  0.07 with SD : 0.17

21

Calculation

P. Sienaert et al. / Journal of Affective Disorders 146 (2013) 15–33

(Lesar et al., 1985)

Lesar et al. (1985)

22

Table 3 (continued ) Predictive method

Articles

Study design

Number of patients

Number of data sets

Outcome parameters

Results Success rate Rate

20.7% 31.0%

Criterium of success or targeted goal

(Terao et al., 1995)

Application

18

18

C

37.9% NM

No deviation o 0.1 mmol/l deviation o 0.2 mmol/l deviation NM

(Terao et al., 1999)

Calculation

30

30

C

20%

o 0.2 mmol/l deviation

(Srisurapanont et al., 2000)

Calculation

17

17

D

NM

NM

(Wright and Crismon, 2000) Calculation (Chiu et al., 2007) Calculation

47 129

47 129

D D

NM NM

NM NM

(Abou-Auda et al., 2008)

Calculation

60

60

D

NM

NM

(Huang et al., 2008)

Calculation

30

30

C

13/30

o 0.2 mEq/l deviation

(Radhakrishnan et al., 2012) Calculation

31

31

C

NM

NM

(Higuchi et al., 1988)

45

219

C

NM

NM

Calculation

*None of the patients showed clinical signs of toxicity *10.3% of the patients had deviations 4 0.2 mmol/l

*Concentrations deviated largely from desired concentration of 0.4 mmol/l: most patients had higher concentrations *4/18 patients had a concentration of 4 1.2 mmol/l *1/18 patients had a concentration o 0.4 mmol/l *Deviations correlated well with measures of renal function, even when these were within normal limits MPE : 0.52 with SD : 0.42 and MAE: 0.54 with SD: 0.39 *MPE: 59.12 with CI: (  15.48;133.72) and RMSE: 152.66 with CI: (81.92;199.75) *In 11 patients dosing errors 4149 mg/d (‘‘predictable error’’) occurred MPE: 170.80 with CI: (81.60;259.99) *Toxicity group : MPE : 108.3 with CI: (39.2;177.4) *Control group: MPE: 101.5 with CI: (21.0;182.1) *For inpatients: MPE: 68.36 with SD: 32.64; MAE: 181.13 with SD: 20.98; RMSE: 229.36 with SD: 110.72; percent error: 12.45% with SD: 4.66 *for outpatients: MPE: 47.79 with SD: 26.68; MAE: 147.24 with SD: 17.55; RMSE: 190.95 with SD: 94.92; percent error: 8.76% with SD: 3.92 *MPE:  0.27 with SD : 0.25 and MAE: 0.29 with SD: 0.39 *8/30 patients had a deviation of 40.4 mEq/l *MPE: 0.46 and RMSE: 0.59 *Error of prediction: 61.33% *Sensitivity in detecting lithium levels outside therapeutic range: 90% *Specificity in detecting lithium levels outside therapeutic range: 71.42% *MPE: 0.069 with CI: (0.045;0.094) and RMSE: 0.219 with CI: (0.194;0.242) *Overprediction negatively correlated to age and serum creatinine: greatest

P. Sienaert et al. / Journal of Affective Disorders 146 (2013) 15–33

Higuchi Mason (Higuchi et al., 1988)

Other values and conclusions of interest

Groves et al. (1991) (Groves et al., 1991)

Jermain NONMEM (Jermain et al., 1991)

23

23

C

20/23

0.6–1.2 mEq/l

(Srisurapanont et al., 2000)

Calculation

17

17

D

NM

NM

(Yukawa et al., 1993)

Calculation

34

60

C D

NM

NM

(Taright et al., 1994)

Calculation

35

35

C

NM

NM

(Wright and Crismon, 2000) Calculation

47

47

C

NM

NM

D

*For concentration: MPE: 0.28 with CI: (0.218;0.343) and MAE: 0.29 with CI: (0.235;0.352) *For dosage: MPE:  204.5 with CI: (  247.6;  161.4) and MAE: 215.2 with CI: (175.9;254.5) Bias:  0.44 Prediction: 0.22 *For concentration : MPE : 0.11 with CI: (0.02;0.19) *For dosage: MPE:  130.41 with CI: (  232.58;  28.25) *For inpatients: MPE:  1.066 with SD: 0.107; MAE: 1.07 with SD: 0.105; RMSE: 1.29 with SD: 0.67; percent error:  46.37% with SD: 2.69 *For outpatients: MPE: -0.902 with SD: 0.068; MAE: 0.91 with SD: 0.07; RMSE: 0.99 with SD: 0.35; percent error:  44.34% with SD: 2.75

Calculation

60

60

Cl

NM

NM

Moscovich et al. (1992)

(Moscovich et al., 1992)

Application

9

9

O

NM

NM

* 4 50% improvement in MSRS (BiegelMurphy Mania State Rating Scale) after 1 week of therapy *After 3 weeks the 3 psychiatric rating scales (MSRS, Clinical Global Inventory, Brief Psychiatric Rating Scale) showed striking improvement and all but one patients were clinically close to discharge *No toxic manifestations were observed

Yukawa NONMEM (Yukawa et al., 1993)

(Yukawa et al., 1993)

Calculation

34

60

C D

NM

NM

(Taright et al., 1994)

Calculation

35

35

D

NM

NM

*For concentration: MPE:  0.02 with CI: (  0.053;0.022) and MAE: 0.11 with CI: (0.089;0.137) *For dosage: MPE: 32.1 with CI: (  13.4;77.5) and MAE: 135.2 with CI: (105.7;164.7) Bias:  0.04 Prediction: 0.01

Sproule Fuzzy logic (Sproule et al., 1997)

(Sproule et al., 1997)

Calculation

10

50

D

NM

NM

MPE: 0.03 with CI: ( 0.01;0.06) and RMSE: 0.13 with CI: (0.09;0.16)

Terao et al. (1999)

(Terao et al., 1999)

Calculation

30

30

C

63%

o 0.2 mmol/l deviation

(Chiu et al., 2007)

Calculation

129

129

D

NM

NM

*MPE : 0.15 with SD: 0.3 and MAE: 0.22 with SD: 0.25 *22/30 patients achieved higher concentrations than expected and 8/30 patients had lower concentrations than expected *Toxicity group : MPE : 251.7 with CI: (156.7;346.7)

23

(Abou-Auda et al., 2008)

P. Sienaert et al. / Journal of Affective Disorders 146 (2013) 15–33

Application

overprediction in patients o 15y or with low serum creatinine level 1/23 patients had a concentration below the range and 2/23 patients above (1.3 and 1.4 mEq/l) *MPE:  179.78 with CI: (  293.85;  65.71) and RMSE: 280.44 with CI: (128.82;375.10) *In 7 patients dosing errors of 4149 mg/ d (‘‘predictable error’’) occurred

24

Table 3 (continued ) Predictive method

Articles

Study design

Number of patients

Number of data sets

Outcome parameters

Results Success rate Rate

Criterium of success or targeted goal

(Abou-Auda et al., 2008)

Calculation

60

60

D

NM

NM

(Huang et al., 2008)

Calculation

30

30

C

8/30

o 0.2 mEq/l deviation

(Keck et al., 2001)

Application

15

15

C

100%

4 0.6 mEq/l after 1 day of treatment

O

(Chiu et al., 2007)

Calculation

129

129

D

NM

NM

Chiu et al. (2007)

(Chiu et al., 2007)

Calculation

129

129

D

NM

NM

Abou-Auda et al. (2008)

(Abou-Auda et al., 2008)

Calculation

60

60

Cl D

NM

NM

(Radhakrishnan et al., 2012) Calculation

31

31

C

NM

NM

*Control group: MPE: 222.9 with CI: (147.5;298.3) *For inpatients: MPE:  265.72 with SD: 33.51; MAE: 282.64 with SD: 30.49; RMSE: 352.85 with SD: 161.56; percent error:  23.13% with SD: 3.39 *for outpatients: MPE:  218.43 with SD: 27.74; MAE: 248.08 with SD: 20.59; RMSE: 280.17 with SD: 108.30; percent error:  18.83 with SD: 3.34 *MPE: 0.31 with SD : 0.22 and MAE: 0.33 with SD: 0.34 *10/30 patients had a deviation of 4 0.4 mEq/l *Mean serum concentration at day 5 was 1.1 þ /- 0.2 mEq/l *60% had 450% reduction on YMRS scores after 10 days and were considered acute responders *2/15 patients terminated prematurely because of adverse effects *3/15 patients developed bradycardia *Toxicity group: MPE: 268.1 with CI: (205.2;330) *Control group: MPE: 267.4 with CI: (171.9;362.9) *Toxicity group : MPE :  39.6 with CI: (  124.6;45.3) *Control group: MPE:  90.2 with CI: (  183.6;3.2) *For clearance in inpatients: MPE: 0.002 with SD: 0.10; MAE: 0.49 with SD: 0.07; RMSE: 0.69 with SD: 0.45; percent error: 10.31% with SD: 5.77 *For clearance in outpatients: MPE: 0.004 with SD: 0.063; MAE: 0.334 with SD: 0.035; RMSE: 0.40 with SD: 0.165; percent error: 6.62% with SD: 4.80 *For dosage in inpatients: MPE: 0.68 with SD: 26.6; MAE: 142.9 with SD: 16.86; RMSE: 184.57 with SD: 95.99; percent error: 3.96% with SD: 3.32 *For dosage in outpatients: MPE: 0.026 with SD: 23.47; MAE: 111.5 with SD: 15.5; RMSE: 148.46 with SD: 72.16; percent error: 2.95% with SD: 3.12 *MPE: 0.24 and RMSE: 0.57 *Error of prediction: 61.52% *Sensitivity in detecting lithium levels outside therapeutic range: 90%

P. Sienaert et al. / Journal of Affective Disorders 146 (2013) 15–33

Keck et al. (2001)

Other values and conclusions of interest

This method was called the ‘method of Gangadhar’ in the article, but seems to correspond to the equation of Sampath. a

NM: not mentioned in the article. Application: the formula was applied to initiate lithium therapy in patients. Calculation: the formula was not applied to initiate therapy, but calculations based on the formula were compared with the actual levels/doses obtained after initiation. D: dosage (in mg/d). C: concentration (in mEq/l or mmol/l). Cl: lithium clearance (in l/h). O: other. MPE: mean prediction error (expressed in the unit mentioned with the outcome parameter studied). MAE: mean absolute prediction error (expressed in the unit mentioned with the outcome parameter studied). RMSE : root mean squared prediction error (expressed in the unit mentioned with the outcome parameter studied). SD: standard deviation. CI : 95% confidence interval. Toxicity group/control group: group of patients with a history of lithium toxicity / group of patients without a history of lithium toxicity.

*MPE: 0.02 with SD : 0.20; MAE: 0.16 with SD: 0.22 *2/30 patients had a deviation of 40.4 mEq/l o 0.2 mEq/l deviation 19/30 C 30 30 Calculation (Huang et al., 2008) Huang (Huang et al., 2008)

*Specificity in detecting lithium levels outside therapeutic range: 66.67%

P. Sienaert et al. / Journal of Affective Disorders 146 (2013) 15–33

25

found poor results with a dosing and sampling interval of 12 h or 36 h, while good results were found for an interval of 24 h. A well-considered choice of sampling time also seems an important factor in the method of Slattery, where it was demonstrated that interindividual differences in pharmacokinetic parameters only minimally affect the predictions when the sampling time can be chosen to be around 16 h (Slattery, 1981; Zimmerman and Slattery, 1983). In a recent study of three a priori methods in Indian patients, Radhakrishnan et al. (2012), found none of the tested predictive methods accurate enough for use in routine clinical practice, making the authors to suggest that the lack of ethnic variables in the predictive methods may hamper their accuracy. An important variable, limiting the generalizability of the data, is the method used to measure the level of lithium in the serum sample. In test-dose methods, such as the Cooper-method, predictions depend on the measured lithium level 24 h after the test dose. Since these 24-hour serum lithium levels are low, correct biochemical analysis is not evident and has to be determined to the second decimal point (Cooper et al., 1973). It is unsure whether this can be done reliably by flame photometry or has to be done by atomic absorption spectrophotometry, the technique used in the original Cooper-study, but which is often not available in the standard biochemical laboratory (Karki et al., 1989). In one study, measuring lithium levels with flame photometry analysis, resulted in a poor performance of the Coopermethod (Palladino et al., 1983). Karki et al. (1989) tested a modified two-point prediction technique and used both flame photometric and atomic absorption spectrophotometric analysis: no significant differences were found. Although the risk of toxicity is one of the major concerns when initiating lithium, only eight of the studies examining predictive methods explicitly report on toxicity (or adverse effects) as an outcome measure (Cummings et al., 1993; Gangadhar et al., 1989a, 1989b; Keck et al., 2001, 1985; Moscovich et al., 1992; Peterse et al., 1999; Wheeler et al., 2008). Noteworthy is the fact that six of these studies specifically examine lithium loading or body weight methods, a category of predictive methods where the risk of toxicity is obvious since these methods aim at directly achieving lithium levels around 1 mEq/l with a starting dosage being only based on the patient’s body weight (Gangadhar et al., 1989a,1989b; Keck et al., 2001; Kook et al., 1985; Moscovich et al., 1992; Wheeler et al., 2008). 4.2. Practical applicability A higher number of test doses and/or blood samples hampers the acceptability of some methods, both for clinicians and patients. Multiple-point methods, such as the Perry multiplepoint method, require two blood samples after the test dose to calculate the lithium dosage that the patient will need for achieving a certain therapeutic lithium level. This implies a time-delay of at least 36 h before therapy can be started (Browne et al., 1988; Lobeck et al., 1987). Moreover, a higher number of serum level measurements increases the risk of assay error (Nelson, 1988). On the other hand, the multiple-point methods allow a greater margin in the choice of sampling time as only the time interval between the samples is fixed, while the interval between the test dose and the first sample is flexible (Karki et al., 1987). A good dosing method should achieve adequate lithium levels as quick as possible. It was shown that manic patients, in which lithium was initiated using the empirical dose titration, spent on average three to four days longer on an acute ward than patients in a ‘modified Slattery’ dosing group (Marken et al., 1994). However, due to the increasing use of atypical antipsychotics in

26

Table 4 Studies examining test-dose predictive methods. Predictive method

Articles

Study design

Number of patients

Number of data sets

Outcome parameters

Results Success rate

Cooper et al. (1973) (Cooper and Simpson, 1976) Application

Other values and conclusions of interest

Rate

Criterium of success or targeted goal

4100

4100

D/C

NM

NM

Application Application

24 13

24 13

D C

23/24 69.3%

0.8–1.1 mEq/l 0.6–1.2mEq/l

(Perry et al., 1983)

Calculation

17

17

D/C

NM

NM

(Palladino et al., 1983) (Fava et al., 1984)

Application Application

17 30

17 30

C C

8/17 80% 96.6%

0.6–1.2 mEq/l 0.6–1.2mEq/l at day 3 0.6–1.2 mEq/l at day 5–6

(Karki et al., 1987)

Calculation

20

20

C

NM

NM

(Browne et al., 1988)

Calculation

20

20

C

50%

o 0.2 mEq/l deviation

(Naganuma et al., 1988)

Application

20

20

C

85% 85%

0.6–1.2 mEq/l 0.6–1.2 mEq/l

(Nelson, 1988)

Simulation

0

950

C

NM

NM

(Kuruvilla and Shaji, 1989)

Application

35

35

C

42.86%

0.6–1.2 mEq/l

(Peterse et al., 1999)

Application

10

10

D

NM

NM

Gaillot et al. (1979) (Gaillot et al., 1979)

Calculation

24

Pharmacokinetic parameters randomly

C

495%

0.8–2 mmol/l

NM

P. Sienaert et al. / Journal of Affective Disorders 146 (2013) 15–33

(Gengo et al., 1980) (Naiman et al., 1981)

Two year follow-up study: predicted dosage requirements have not required alteration NM 30.7% of patients had concentrations outside the therapeutic range: 3 patients had subtherapeutic concentrations and 1 patient exceeded the therapeutic range No significant difference between dosage recommendations from Cooper and Perry ‘83 single point or between theoretical steady state levels predicted by these 2 methods NM *Number of days and serum lithium determinations needed to achieve therapeutic steady-state levels: 5.53 þ/  1.13 and 2,2 þ/  0.4 respectively *Mean length of stay: 17.26 þ /  8.86 significant overprediction of concentrations *Median prediction error : 0.17 with CI: (0.00;0.38) *MAE: 0.01 3/20 patients had subtherapeutic concentrations *For SD 5%: MPE:  0.012 with SD: 0.153 and CI: ( 0.022;-0.002); mean squared prediction error (mEq/l)2: 0.024 with SD: 0.129 and CI: (0.016– 0.032) *For SD10%: MPE:  0.016 with SD: 0.176 and CI: ( 0.027;  0.005); mean squared prediction error (mEq/l)2: 0.031 with SD: 0.149 and CI: (0.021–0.041) *Prediction errors are correlated to the elimination rate constant: larger errors are made in patients with long half-lives The other 57.14% of patients required additional lithium doses May lead to potentially toxic lithium dose recommendations, especially when very low serum levels are measured after the test dose

Application

40

(Palladino et al., 1983)

Calculation

17

17

D

10/17

o 20% deviation of actual dosage

(Lobeck et al., 1987)

Calculation

16

16

C

NM

NM

(Valecha et al., 1990)

Calculation

46

46

D

11/46

o 150 mg/d deviation

Tyrer et al. (1981)

(Tyrer et al., 1981)

Calculation

24

24

C

NM

NM

Norman et al. (1982) Perry et al. (1982)

(Norman et al., 1982)

Application

15

17

C

NM

NM

(Perry et al., 1982)

Calculation

22

22

C

91%

23% 55%

o 0.1 mEq/l deviation for the time points *24, 36 h *12, 24, 36 h *12, 36 h *12, 24 h o 0.1 mEq/l deviation o 0.15 mEq/l deviation NM

Slattery et al. (1980)

88–93%

0.6–0.9 mmol/l

Overprediction of dosage in 1/40 patients and underprediction of dosage in 2/40 patients A strong positive correlation was found between the reciprocal of the maintenance dose required to achieve a steady state concentration of 1.0 mEq/l and the 24 h serum lithium level, as described by Slattery MPE: 0.149 with CI: (0.028;0.270) and Mean squared error (mEq/l)2: 0.071 with CI (  0.014;0.627) *9/46 patients had a deviation of 150– 300 mg/d, 8/46 patients a deviation of 300–600 mg/d and in 18/46 patients a deviation of 4 600 mg/d occured *30/46 patients had higher predictions than observed and 5/46 patients had lower predictions than observed

91%

(Perry et al., 1986)

Application

30

30

C

(Karki et al., 1987)

Application

5

5

C

77% 80% NM

(Browne et al., 1988)

Calculation

20

20

C

67%

o 0.2 mEq/l deviation

(Williams et al., 1989)

Calculation

21

21

C

NM

NM

*Lithium clearance and the 10 or 17 h serum lithium value after a single dose are accurate and comparable in the prediction of steady state levels *The published formulae are only indicative and are not set forward as the final predictive equations Average prediction error: 5.2% (range—13% to 9%) NM

MPE:  0.047 mean squared error: 0.029(mEq/l)2 significant underprediction of concentrations *Median prediction error:  0.06 with CI: (  0.19;0.01) and MAE: 0.06 *in 5/6 patients with 40.2 mEq/l deviation the concentration was underpredicted Median prediction error:  0.015 with CI: (  0.075;0.035) and MAE: 0.10

(Marr et al., 1983)

Application

6

6

C

5/6

o 0.1 mEq/l deviation

MPE: 0.07 with SD : 0.08

(Perry et al., 1983)

Calculation

17

17

D/C

NM

NM

(Perry et al., 1984)

Application

38

38

C

83%

0.4–0.89 mEq/l (prophylactic range) 0.9–1.3 mEq/l (range for acute mania)

No significant difference between dosage recommendations from Cooper and Perry ‘83 single point or between theoretical steady state levels predicted by these 2 methods NM

85%

27

Marr et al., 1983) Perry et al. (1983)

P. Sienaert et al. / Journal of Affective Disorders 146 (2013) 15–33

(Klinger et al., 1984)

selected from the study popula-tion 40 C

28

Table 4 (continued ) Predictive method

Articles

Study design

Number of patients

Number of data sets

Outcome parameters

Results Success rate

(Perry et al., 1984)

Calculation

32

49 (also data from Perry ‘83 included)

D

Perry et al. (1986)

(Perry et al., 1986)

Calculation

30

30

C

Rate

Criterium of success or targeted goal

NM

NM

o 0.1 / o0.15 mEq/l deviation for the time point(s)

(Browne et al., 1988)

Calculation

20

20

C

40%/53% 43%/73% 60%/70% 85%

(Williams et al., 1989)

Calculation

21

21

C

NM

NM

(Browne et al., 1989)

Calculation

20

20

C

85%

o 0.2 mEq/l deviation

(Gervasoni et al., 2003)

Application

31

31

C

51.6%

0.7–0.9 mmol/l

Karki Corrected multiple point of Perry ‘82 (Karki et al., 1987)

(Karki et al., 1987)

application

15

15

C

NM

NM

*No significant difference between observed and predicted concentrations *Best results with 12 h and 36 h sampling time points

Lobeck Modified Slattery (Lobeck et al., 1987)

(Lobeck et al., 1987)

calculation

16

16

C

NM

NM

(Marken et al., 1994)

application

18

18

C

1/6

o 0.2 mEq/l deviation

MPE:  0.042 with CI: ( 0.158;0.074) and mean squared error (mEq/l)2: 0.046 with CI: (0.015;0.078) 5/6 patients had deviations 40.2 mEq/l: for 1 patient the dosage was too high and for 4 patients the dosage was too low

Calculation Calculation

19 46

19 46

C D

17/19 NM

0.6–1.2 mEq/l NM

NM In the majority of patients an overprediction of dosage occured, especially in the patients with low levels after the test dose

Calculation

45

283

C

NM

NM

*Using Pepin population mean values: MPE: 0.025 with CI: (0.015;0.035) and RMSE: 0.219 with CI: (0.205;0.232) *Using Mason population mean values: MPE: 0.015 with CI: (0.005;0.024) and RMSE: 0.213 with CI: (0.200;0.226)

Rosenberg Modified (Rosenberg et al., 1987) (Valecha et al., 1990) Cooper (Rosenberg et al., 1987) (Higuchi et al., 1988)

Higuchi Bayesian (Higuchi et al., 1988)

*12 h *24 h *36 h o 0.2 mEq/l deviation

Modification of Perry ‘83 method: improvement in coefficient of determination, but no dramatic improvement in observed dosing scheme’s accuracy, based on recalculation of the doses for the patients who failed to achieve desired concentration based on the original equations *for 12 h: MPE:  0.143 and mean squared error (mEq/l)2: 0.054 *for 24 h: MPE:  0.053 and mean squared error (mEq/l)2: 0.026 *for 36 h: MPE: -0.090 and mean squared error (mEq/l)2: 0.034 Median prediction error:  0.05 with CI: (  0.07;0.02) and MAE: 0.05 Median prediction error:  0.040 with CI: (  0.110;0.015) and MAE: 0.12 Median prediction error : 0.04 with CI: (  0.02;0.07) 22.6% of patients had levels o 0.7 mmol/l and 25.8% had levels 40.9 mmol/l 1/32 patient had a toxic level of 1.32 mmol/l, without clinical signs of toxicity

P. Sienaert et al. / Journal of Affective Disorders 146 (2013) 15–33

Perry et al. (1984)

Other values and conclusions of interest

(Higuchi et al., 1988)

Calculation

45

283

C

NM

NM

*Using equation I: MPE: 0.031 with CI: (0.008;0.054) and RMSE: 0.503 with CI: (0.312;0.705) *Using equation II: MPE: 0.132 with CI: (0.092;0.173) and RMSE: 0.873 with CI: (0.698;1.018)

Nelson (1988)

(Nelson, 1988)

Simulation

0

950

C

NM

NM

*With 12 h interval, 5% SD: MPE: 1.425 with SD: 45.055 and CI: ( 1.44;1.290); mean squared prediction error (mEq/l)2: 2029.826 with SD: 58824.446 and CI: (  1749.038;5808.690) *With 12 h interval, 10% SD: MPE:  0.116 with SD: 28.135 and CI: (  1.905;1.673); mean squared prediction error (mEq/l)2: 790.733 with SD: 18264.610 and CI: (  382.580;1964.046) *With 24 h interval, 5% SD: MPE: 0.081 with SD: 0.256 and CI: (0.065;0.097); mean squared prediction error (mEq/l)2: 0.072 with SD: 0.894 and CI: (0.015;0.129) *with 24h interval, 10% SD: MPE:  0.182 with SD: 9.187 and CI: (  0.766;0.402); mean squared prediction error (mEq/l)2: 84.339 with SD: 1820.436 and CI: (  32.322;201.000)

Williams Bayesian (Williams et al., 1989)

(Williams et al., 1989)

Calculation

21

21

C

NM

NM

*0, 12, 24, 36 h time points: median prediction error: -0.055 with CI: (  0.155;0.010) and MAE: 0.08 *0, 12, 24 h time points: median prediction error:  0.045 with CI: (  0.125;0.040) and MAE: 0.13 *0, 12, 36 h time points: median prediction error:  0.030 with CI: (  0.095;0.045) and MAE: 0.07 *0, 24, 36 h time points: median prediction error:  0.045 with CI: (  0.125;0.020) and MAE: 0.09 *0, 12 h time points: median prediction error:  0.020 with CI: (  0.065;0.125) and MAE: 0.15 *0, 24 h time points: median prediction error:  0.055 with CI: (  0.130;0.055) and MAE: 0.13

Williams Non-linear (Williams et al., 1989) regression (Williams et al., 1989)

Calculation

21

21

C

NM

NM

*0, 12, 24, 36 h time points: median prediction error:  0.025 with CI: (-0.120;0.060) and MAE: 0.13 *0, 12, 24 h time points: median prediction error:  0.065 with CI: (  0.160;0.055) and MAE: 0.16 *0. 12, 36 h time points: median prediction error:  0.045 with CI: (  0.135;0.020) and MAE: 0.10 *0. 24. 36 h time points: median prediction error:  0.0 with CI: (  0.085;0.075) and MAE: 0.11

P. Sienaert et al. / Journal of Affective Disorders 146 (2013) 15–33

Higuchi Fixed volume of distribution iterative method (Higuchi et al., 1988)

29

30

Table 4 (continued ) Predictive method

Articles

Study design

Number of patients

Number of data sets

Outcome parameters

Results Success rate

Other values and conclusions of interest

Rate

Criterium of success or targeted goal

(Valecha et al., 1990)

Calculation

46

46

D

27/46

o 150 mg/d deviation

Swartz (1991)

(Swartz, 1991)

Calculation

0

0

O

NM

NM

Taright Bayesian (Taright et al., 1994)

(Taright et al., 1994)

Calculation

35

35

C

NM

NM

*11/46 patients had a deviation of 150– 300 mg/d and in 8/46 patients a deviation 4 300 mg/d occured *12/46 patients had higher predictions than observed and 7/46 patients had lower predictions than observed Theoretical replication of Cooper’s data: comparable results were found for predicted concentrations *Using covariates only: bias  0.06, prediction 0.09 (mmol/l) *using covariates and 24 h concentration after a test dose: bias  0.01, prediction 0.05 (mmol/l) *A serious risk of deviation of the steady state concentration from the therapeutic range was adequately predicted in 4/5 patients in whom it was observed. In 8 patients a risk was mistakenly predicted.

NM: not mentioned in the article. Application: the formula was applied to initiate lithium therapy in patients. Calculation: the formula was not applied to initiate therapy, but calculations based on the formula were compared with the actual levels/doses obtained after initiation. Simulation: the formula was applied on computer simulated subjects, with normally distributed population pharmacokinetic parameters and the introduction of an assay and timing error with mean zero and standard deviation of 5 or 10%. D: dosage (in mg/d). C: concentration (in mEq/l or mmol/l). O: other. MPE: mean prediction error (expressed in the unit mentioned with the outcome parameter studied). MAE: mean absolute prediction error (expressed in the unit mentioned with the outcome parameter studied). RMSE: root mean squared prediction error (expressed in the unit mentioned with the outcome parameter studied). SD: standard deviation. CI: 95% confidence interval.

P. Sienaert et al. / Journal of Affective Disorders 146 (2013) 15–33

Valecha et al. (1990)

P. Sienaert et al. / Journal of Affective Disorders 146 (2013) 15–33

the acute management of severe manic or mixed episodes, this time lapse before achieving a therapeutic lithium level has become less important (APA, 2002). Some of the methods (Cooper et al., 1973; Gaillot et al., 1979; Groves et al., 1991; Keck et al., 2001; Kook et al., 1985; Lesar et al., 1985; Moscovich et al., 1992; Perry et al., 1984; Perry et al., 1983; Rosenberg et al., 1987; Sampath et al., 1981; Valecha et al., 1990) only provide the dosages needed for achieving a preset concentration or concentration range. The Cooper-method (Cooper et al., 1973) and the modified Cooper-method (Rosenberg et al., 1987), for example, target a level between 0.6–1.2 mEq/l; the method of Valecha et al. (1990) targets a lithium level of 0.9 mEq/l. It is not possible for the clinician to adjust these target concentration (ranges). Also, some methods aim specifically at application with slow release preparations (Keck et al., 2001; Kook et al., 1985; Lesar et al., 1985) or rapid release preparations (Karki et al., 1987). The dosing interval (i.e. multiple daily dosing, once-daily dosing) is generally not taken into account as a major factor in the performance of the methods. It has recently been demonstrated that differences in administration schemes do not lead to significant differences in plasma lithium concentrations, although a reduction in daily dose can probably be effectuated when lithium is administrated only once daily (Malhi and Tanious, 2011). The vast majority of predictive methods reviewed here appear to show inconsistent or rather poor results in the studies examining them. Even though a systematic scrutiny of the methodology used in the studies is beyond the scope of the present article, methods that do not perform adequately in the tests should not be recommended in clinical practice (e.g. the methods of Zetin et al. (1983,1986), Pepin et al. (1980), Keck et al. (2001), Cooper et al. (1973), Slattery et al. (1980)). A number of methods have been proposed, but have never been replicated (e.g. the methods of Chiu et al. (2007), Marr et al. (1983), Valecha et al. (1990), Moscovich et al. (1992)). Due to a lack of consistent reporting of results in the various studies reviewed, a statistical analysis or meta-analysis was impossible. The articles use different outcome parameters (e.g. lithium level versus lithium dosage) and different outcome measures for these parameters (e.g. mean absolute error versus root mean squared error) and often do not provide sufficient data to make a pooled analysis possible. Although some of the methods are potentially useful, the lack of sufficient data on their performance, precludes their recommendation in clinical practice: the methods are either insufficiently studied or show rather poor results and therefore may not provide any relevant advantage over the clinical titration method. Surprisingly, the titration method, although being the most widely used, has not been subjected to rigorous study. The only empirical results regarding this method were derived from studies in which this method was used as a control for other predictive methods (Abou-Auda et al., 2008; Browne et al., 1989; Fava et al., 1984; Gangadhar et al., 1989a; Marken et al., 1994; Markoff and King, 1992; Wright and Crismon, 2000). Drawing conclusions about the titration method is furthermore hampered by the fact that ‘the’ titration method does not exist: each clinician adapts this method according to his own clinical feeling and experience. While the performance of the predictive methods varies widely between the articles studied, thus hampering a general recommendation of what method to use, the titration method almost invariably produces reasonable results. The method proved to be equal to the predictive methods, or, was shown to be more conservative, leading to an underestimation of dosage and a longer time to reach therapeutic lithium levels, but as well carrying less risks of adverse effects or toxicity. This was demonstrated by Wright and Crismon (2000), who concluded there was, apart from a tendency to underdosing with the empirical method, no difference in ability to predict lithium

31

dosages between three a priori dosing methods and the empirical method. Likewise, Wheeler et al. (2008) found no difference in clinical outcome measures between patients started with lithium titration or loading, while more adverse effects were reported with the loading strategy (63.6% versus 38.7%). The time lapse between the first dose and an adequate lithium level is probably the major and only disadvantage of the empirical titration method. As noted previously, this time lapse before achieving therapeutic lithium levels had become of minor importance. Based upon the available data, we recommend to use the empirical titration method for initiation of lithium therapy in a standard way in clinical practice. 4.3. Limitations In the present review, the methodological strength was not taken into account for a study to be included.

5. Conclusion The empirical titration method is the most widely used method for lithium initiation. It is a time-honored method that has proven its safety and reliability in clinical practice. Although it is not a fast method, ‘faster’ methods, e.g. the body weight/ loading dose methods, remain understudied and cannot be recommended for clinical use. Based upon the available data, we recommend to use the empirical titration method for initiation of lithium therapy in a standard way in clinical practice. However, regular monitoring of lithium levels remains essential. Role of funding source This study was performed without external funding sources.

Conflict of interest The authors declare no conflicts of interest concerning this study.

Acknowledgments The authors would like to thank Mr. M. Stroobants, who assisted with locating and retrieving the full-text papers.

References Abou-Auda, H.S., Al-Yamani, M.J., Abou-Shaaban, R.R., Khoshhal, S.I., 2008. A new accurate method for predicting lithium clearance and daily dosage requirements in adult psychiatric patients. Bipolar Disorders 10, 369–376. American Psychiatric Association, 2002. Practice guideline for the treatment of patients with bipolar disorder. The American Journal of Psychiatry, 1–50. ¨ Bauer, M., Grof, P., Muller-Oerlingausen, B., 2006. Lithium in neuropsychiatry: the comprehensive guide. Informa healthcare, London. Browne, J.L., Huffman, C.S., Golden, R.N., 1989. A comparison of pharmacokinetic versus empirical lithium dosing techniques. Therapeutic Drug Monitoring 11, 149–154. Browne, J.L., Patel, R.A., Huffman, C.S., Hussey, B.K., 1988. Comparison of pharmacokinetic procedures for dosing lithium based on analysis of prediction error. Drug Intelligence & Clinical Pharmacy 22, 227–231. Bryant, S.G., Guernsey, B.G., Hokanson, J.A., Doutre, W.H., 1984. Predicting lithium dose. The Journal of Clinical Psychiatry 45, 95. Cade, J.F., 1949. Lithium salts in the treatment of psychotic excitement. The Medical Journal of Australia 2, 349–352. Chiu, C.C., Shen, W.W., Chen, K.P., Lu, M.L., 2007. Application of the Cockcroft–Gault method to estimate lithium dosage requirement. Psychiatry and Clinical Neurosciences 61, 269–274. Cooper, T.B., Bergner, P.E., Simpson, G.M., 1973. The 24-hour serum lithium level as a prognosticator of dosage requirements. The American Journal of Psychiatry 130, 601–603. Cooper, T.B., Simpson, G.M., 1976. The 24-hour lithium level as a prognosticator of dosage requirements: a 2-year follow-up study. The American Journal of Psychiatry 133, 440–443. Cummings, M.A., Haviland, M.G., Cummings, K.L., Wareham, J.G., 1988. Lithium dose prediction: a prospective case series. The Journal of Clinical Psychiatry 49, 373.

32

P. Sienaert et al. / Journal of Affective Disorders 146 (2013) 15–33

Cummings, M.A., Haviland, M.G., Wareham, J.G., Fontana, L.A., 1993. A prospective clinical evaluation of an equation to predict daily lithium dose. The Journal of Clinical Psychiatry 54, 55–58. Dugas, J.E., Feeney, A.M., 1983. Evaluation of a new lithium dosage-prediction technique. Clinical Pharmacy 2, 249–252. Fava, G.A., Molnar, G., Block, B., Lee, J.S., Perini, G.I., 1984. The lithium loading dose method in a clinical setting. The American Journal of Psychiatry 141, 812–813. Gaillot, J., Steimer, J.L., Mallet, A.J., Thebault, J.J., Bieder, A., 1979. A priori lithium dosage regimen using population characteristics of pharmacokinetic parameters. Journal of Pharmacokinetics and Biopharmaceutics 7, 579–628. Gangadhar, B.N., Mahajan, D.S., Narasimha Reddy, P., Rao, Sridhara Rama, Janakiramaiah, N., B.S., 1989a. Predicting lithium dose on the basis of body weight to achieve rapid therapeutic levels. Indian Journal of Psychological Medicine 12, 7–10. Gangadhar, B.N., Narasimha Reddy, P., Mahajan, D.S., Rao, Sridhara Rama, Janakiramaiah, N., B.S., 1989b. More on prediction of lithium levels or concentrations. Biological Psychiatry 26, 214–218. Gengo, F., Timko, J., D’Antonio, J., Ramsey, T.A., Frazer, A., Mendels, J., 1980. Prediction of dosage of lithium carbonate: use of a standard predictive method. The Journal of Clinical Psychiatry 41, 319–320. Gervasoni, N., Zona-Favre, M.P., Osiek, C., Roth, L., Bondolfi, G., Bertschy, G., 2003. Lithium dose prediction based on 24 h single dose levels: a prospective evaluation. Pharmacological Research 48, 649–653. Grandjean, E.M., Aubry, J.M., 2009. Lithium: updated human knowledge using an evidence-based approach. Part II: Clinical pharmacology and therapeutic monitoring. CNS Drugs 23, 331–349. Groves, G.E., Clothier, J.L., Hollister, L.E., 1991. Predicting lithium dose by the bodyweight method. International Clinical Psychopharmacology 6, 19–23. Higuchi, S., Fukuoka, R., Aoyama, T., Horioka, M., 1988. Predicting serum lithium concentration using Bayesian method: a comparison with other methods. Journal of Pharmacobio-dynamics 11, 158–174. Huang, H.C., Chang, Y.L., Lan, T.H., Chiu, H.J., Liu, W.M., Lee, T.J., 2008. Prediction of optimal lithium doses for Taiwanese psychiatric patients. Journal of Clinical Pharmacy and Therapeutics 33, 115–121. Jermain, D.M., Crismon, M.L., Martin 3rd, E.S., 1991. Population pharmacokinetics of lithium. Clinical Pharmacy 10, 376–381. Karki, S.D., Carson, S.W., Holden, J.M., 1989. Effect of assay methodology on the prediction of lithium maintenance dosage. DICP 23, 372–375. Karki, S.D., Carson, S.W., Holden, J.M., Nanavati, D., 1987. Evaluation of a two-point method for prediction of lithium maintenance dosage. International Clinical Psychopharmacology 2, 343–351. Keck Jr., P.E., Strakowski, S.M., Hawkins, J.M., Dunayevich, E., Tugrul, K.C., Bennett, J.A., McElroy, S.L., 2001. A pilot study of rapid lithium administration in the treatment of acute mania. Bipolar Disorders 3, 68–72. Klinger, E., Steimer, J.L., Jouvent, R., Le Moel, G., Mascart, J.Y., Des Lauriers, A., 1984. Renal clearance technic for individualizing lithium dosage in routine hospital care. L’Encephale 10, 223–230. Kook, K.A., Stimmel, G.L., Wilkins, J.N., Spangher, G.G., 1985. Accuracy and safety of a priori lithium loading. The Journal of Clinical Psychiatry 46, 49–51. Kuruvilla, K., Shaji, K.S., 1989. How reliable is 24 h serum lithium level after a test dose of lithium in predicting optimal lithium dose? Indian Journal of Psychiatry 31, 70–72. Lauritsen, B.J., Mellerup, E.T., Plenge, P., Rasmussen, S., Vestergaard, P., Schou, M., 1981. Serum lithium concentrations around the clock with different treatment regimens and the diurnal variation of the renal lithium clearance. Acta Psychiatrica Scandinavica 64, 314–319. Lesar, T.S., Tollefson, G., Koch, M., 1985. Relationship between patient variables and lithium dosage requirements. The Journal of Clinical Psychiatry 46, 133–136. Lobeck, F., 1988. A review of lithium dosing methods. Pharmacotherapy 8, 248–255. Lobeck, F., Nelson, M.V., Evans, R.L., Hornstra, R.K., 1987. Evaluation of four methods for predicting lithium dosage. Clinical Pharmacy 6, 230–233. Lu, M.L., Shen, W.W., Chen, K.P., 2002. Estimation of lithium dose requirement by body weight. Bipolar Disorders 4, 274. Malhi, G.S., Tanious, M., 2011. Optimal frequency of lithium administration in the treatment of bipolar disorder: clinical and dosing considerations. CNS Drugs 25, 289–298. Malhi, G.S., Tanious, M., Das, P., Berk, M., 2012. The science and practice of lithium therapy. The Australian and New Zealand Journal of Psychiatry 46, 192–211. Malhi, G.S., Tanious, M., Gershon, S., 2011. The lithiumeter: a measured approach. Bipolar Disorders 13, 219–226. Marken, P.A., McCrary, K.E., Lacombe, S., Sommi, R.W., Hornstra Jr., R.K., Pierce, C.A., Stanislav, S.W., Evans, R.L., 1994. Preliminary comparison of predictive and empiric lithium dosing: impact on patient outcome. The Annals of Pharmacotherapy 28, 1148–1152. Markoff, R.A., King Jr., M., 1992. Does lithium dose prediction improve treatment efficiency? Prospective evaluation of a mathematical method. Journal of Clinical Psychopharmacology 12, 305–308. Marr, M.A., Djuric, P.E., Ritschel, W.A., Garver, D.L., 1983. Prediction of lithium carbonate dosage in psychiatric inpatients using the repeated one-point method. Clinical Pharmacy 2, 243–248. Moscovich, D.G., Shapira, B., Lerer, B., Belmaker, R.H., 1992. Rapid lithiumization in acute manic patients. Human Psychopharmacology 7, 343–345. Naganuma, H., Akiyoshi, J., Fujii, I., 1988. Prediction of optimal lithium dosage. Journal of Clinical Psychopharmacology 8, 66–67.

Naiman, I.F., Muniz, C.E., Stewart, R.B., Yost, R.L., 1981. Practicality of a lithium dosing guide. The American Journal of Psychiatry 138, 1369–1371. Nelson, M.V., 1988. Comparison of three lithium dosing methods in 950 ‘‘subjects’’ by computer simulation. Therapeutic Drug Monitoring 10, 269–274. Norman, K.P., Cerrone, K.L., Reus, V.I., 1982. Renal lithium clearance as a rapid and accurate predictor of maintenance dose. The American Journal of Psychiatry 139, 1625–1626. Palladino Jr., A., Longenecker, R.G., Lesko, L.J., 1983. Lithium test-dose methodology using flame photometry: problems and alternatives. The Journal of Clinical Psychiatry 44, 7–9. Patrias, J.M., Moore, F.H., 1985. Evaluation of two methods for predicting initial lithium dosage without measuring serum concentrations. Clinical Pharmacy 4, 77–79. Pepin, S.M., Baker, D.E., Nance, K.S., 1980. Lithium dosage calculation from age, sex, height, weight and serum creatinine. In: Proceedings of the 15th Annual ASHP Midyear Clinical Meeting, San Francisco. Perry, P.J., Alexander, B., Dunner, F.J., Schoenwald, R.D., Pfohl, B., Miller, D., 1982. Pharmacokinetic protocol for predicting serum lithium levels. Journal of Clinical Psychopharmacology 2, 114–118. Perry, P.J., Alexander, B., Prince, R.A., Dunner, F.J., 1984. Prospective evaluation of two lithium maintenance dose schedules. Journal of Clinical Psychopharmacology 4, 242–246. Perry, P.J., Alexander, B., Prince, R.A., Dunner, F.J., 1986. The utility of a single-point dosing protocol for predicting steady-state lithium levels. British Journal of Psychiatry 148, 401–405. Perry, P.J., Prince, R.A., Alexander, B., Dunner, F.J., 1983. Prediction of lithium maintenance doses using a single point prediction protocol. Journal of Clinical Psychopharmacology 3, 13–17. Peterse, J.W., Havenaar, J.M., van Rijn, H.J., 1999. Possible hazard in use of priming dose to determine lithium dosage. The American Journal of Psychiatry 156, 157–158. Poust, R.I., Mallinger, A.G., Mallinger, J., Himmelhoch, J.M., Hanin, I., 1976. Pharmacokinetics of lithium in human plasma and erythrocytes. Psychopharmacology Communications 2, 91–103. Radhakrishnan, R., Kanigere, M., Menon, J., Calvin, S., Srinivasan, K., 2012. Comparison of three a-priori models in the prediction of serum lithium concentration. Indian Journal of Pharmacology 44, 234–237. Ritschel, W.A., Thompson, G.A., 1979. The one-point method in predicting dosage regimen in case of hepatic and/or renal failure in presence or absence of change in volume of distribution. Journal of Clinical Pharmacology 19, 350–356. Rosenberg, J.G., Binder, R.L., Berlant, J., 1987. Prediction of therapeutic lithium dose: comparison and improvement of current methods. The Journal of Clinical Psychiatry 48, 284–286. Sampath, G., Channabasavanna, S.M., Rao, Sridhara Rama, Subhash, M.N., B.S., 1981. Body weight–dose relationship of lithium carbonate in manics. The Indian Journal of Medical Research 74, 750–753. Sheiner, L.B., Beal, S.L., 1981. Some suggestions for measuring predictive performance. Journal of Pharmacokinetics and Biopharmaceutics 9, 503–512. Siemsen, F.K., Pasley, D.R., 1988. Comment: dosing lithium. Drug Intelligence & Clinical Pharmacy 22, 1005. Slattery, J.T., 1981. Single-point maintenance dose prediction: role of interindividual differences in clearance and volume of distribution in choice of sampling time. Journal of Pharmaceutical Sciences 70, 1174–1176. Slattery, J.T., Gibaldi, M., Koup, J.R., 1980. Prediction of maintenance dose required to attain a desired drug concentration at steady-state from a single determination of concentration after an initial dose. Clinical Pharmacokinetics 5, 377–385. Sproule, B.A., Bazoon, M., Shulman, K.I., Turksen, I.B., Naranjo, C.A., 1997. Fuzzy logic pharmacokinetic modeling: application to lithium concentration prediction. Clinical Pharmacology and Therapeutics 62, 29–40. Srisurapanont, M., Pratoomsri, W., Maneeton, N., 2000. Evaluation of three simple methods for predicting therapeutic lithium doses. Psychiatry Research 94, 83–88. Stip, E., Dufresne, J., Boulerice, B., Elie, R., 2001. Accuracy of the Pepin method to determine appropriate lithium dosages in healthy volunteers. Journal of Psychiatry and Neuroscience 26, 330–335. Swartz, C.M., 1984. Predicting lithium requirements. The Journal of Clinical Psychiatry 45, 95–96. Swartz, C.M., 1991. Drug dose prediction with flexible test doses. Journal of Clinical Pharmacology 31, 662–667. Taright, N., Mentre, F., Mallet, A., Jouvent, R., 1994. Nonparametric estimation of population characteristics of the kinetics of lithium from observational and experimental data: individualization of chronic dosing regimen using a new Bayesian approach. Therapeutic Drug Monitoring 16, 258–269. Terao, T., Oga, T., Nozaki, S., Ota, A., Otsubo, Y., Yamamoto, S., Zamami, M., Okada, M., 1995. A further prospective evaluation of an equation to predict daily lithium dose. The Journal of Clinical Psychiatry 56, 193–195. Terao, T., Okuno, K., Okuno, T., Nakano, H., Shinkai, T., Ohmori, O., Yoshimura, R., Suzuki, T., Abe, K., 1999. A simpler and more accurate equation to predict daily lithium dose. Journal of Clinical Psychopharmacology 19, 336–340. Tyrer, S.P., Grof, P., Kalvar, M., Shopsin, B., 1981. Estimation of lithium dose requirement by lithium clearance, serum lithium and saliva lithium following a loading dose of lithium carbonate. Neuropsychobiology 7, 152–158. Valecha, N., Tayal, G., Tripathi, K.D., 1990. Single dose pharmacokinetics of lithium & prediction of maintenance dose in manic depressive patients. The Indian Journal of Medical Research 92, 409–416.

P. Sienaert et al. / Journal of Affective Disorders 146 (2013) 15–33

Wheeler, A., Robinson, G., Fraser, A., 2008. Mood stabilizer loading versus titration in acute mania: audit of clinical practice. The Australian and New Zealand Journal of Psychiatry 42, 955–962. Williams, P.J., Browne, J.L., Patel, R.A., 1989. Bayesian forecasting of serum lithium concentrations. Comparison with traditional methods. Clinical Pharmacokinetics 17, 45–52. Wright, R., Crismon, M.L., 2000. Comparison of three a priori methods and one empirical method in predicting lithium dosage requirements. American Journal of Health-System Pharmacy 57, 1698–1702. Yukawa, E., Nomiyama, N., Higuchi, S., Aoyama, T., 1993. Lithium population pharmacokinetics from routine clinical data: role of patient characteristics for estimating dosing regimens. Therapeutic Drug Monitoring 15, 75–82.

33

Zetin, M., Garber, D., Cramer, M., 1983. A simple mathematical model for predicting lithium dose requirement. The Journal of Clinical Psychiatry 44, 144–145. Zetin, M., Garber, D., De Antonio, M., Schlegel, A., Feureisen, S., Fieve, R., Jewett, C., Reus, V., Huey, L.Y., 1986. Prediction of lithium dose: a mathematical alternative to the test-dose method. The Journal of Clinical Psychiatry 47, 175–178. Zetin, M., Shokooh, S., Daghighi-Asli, A., Plon, L., 1990. Prediction of lithium dose: a follow-up study. Lithium 1, 182–183. Zimmerman, C.L., Slattery, J.T., 1983. Maintenance-dose prediction based on a single determination of concentration: general applicability to twocompartment drugs with reference to lithium. Journal of Pharmaceutical Sciences 72, 1262–1266.