The first level of difficulty in evaluating lithium response is the complexity of bipolar disorder and its varied evolution: instability between bipolar I and bipolar II diagnosis, the unpredictability of the course of the illness (intensity of episodes, frequency of episodes), the induced mood swings, the presence of rapid cycling and psychiatric and physical comorbidities (Malhi et al. 2012).

A second level of difficulty is the extension of the bipolar spectrum to encompass a broad spectrum of disorders that are probably too far from the core to be considered as manic-depressive illness. Recent reports of reduced lithium efficacy might be attributable to the expansion of the diagnostic criteria of bipolar disorder in the Diagnostic and Statistical Manual-based systems (Baldessarini and Tondo 2000; Malhi and Porter 2014). The incorporation of too widely varied phenotypes of mood disorders may have skewed many trials on bipolar disorders, especially those evaluating therapeutic response.

Taking this into account, together with the first level of difficulty regarding complexity, does lithium resistance indeed exist? For some authors (with considerable clinical experience of bipolar disorder and practical use of lithium prescription), once the clinical profile of a good responder is established, there is a ‘written guarantee’ that lithium will provide good stabilization for years (Grof 2006). On the other hand, other authors describe secondary lithium resistance after years of prophylactic efficacy: more severe and/or more frequent breakthrough episodes appear progressively, a pattern consistent with the emergence of tolerance (Post 2012). Among lithium-refractory patients studied (Post 2010), 13.6 % showed this phenomenon after being on lithium for an average of 6.6 years.

A third aspect for defining lithium response is the various therapeutic effects of lithium itself. Beyond the prophylactic effect, lithium has ‘curative’ properties for episodes, as well as clear clinical effects on suicide (Baldessarini et al. 2006) and impulsive behaviours. Each therapeutic property may be independent from one another. Lithium has proven useful in major depressive disorders, particularly with the treatment of resistant depression. To date, lithium’s prophylactic and anti-suicidal effects are unique, and lithium differs from other mood stabilizers, as it reduces the risk of suicide not only through the prevention of mood episodes but also in lithium nonresponders, perhaps due to a specific anti-suicidal mechanism of action. This chapter focuses exclusively on the response to long-term treatment of patients with recurrent mood disorders.

In this respect, a fourth level of difficulty for defining lithium response becomes apparent. Prescription trends in the management of bipolar disorder have shifted away from monotherapy (even if this remains the patient’s preference and the physician’s aim) to combination therapy and polypharmacy. The trend is generally to use combinations of different putatitive mood stabilizers. The treatment of bipolar disorders for 4,700 patients in The Health Improvement Network (THIN) primary was analysed for over 15 years, between 1995 and 2009 (Hayes et al. 2011). In 1995 23 % were issued with two or more prescriptions for more than one psychotropic medication; by 2009 this had increased to 48 %. Seventeen percent of patients were prescribed lithium and an antipsychotic. Moreover, in 1995 5 % were prescribed an anticonvulsant (valproate, carbamazepine or lamotrigine) and an antipsychotic; by 2009 this had increased to 31 %. Lithium and an anticonvulsant were prescribed to 5 % of the population in 1995, and by 2009 this reached 12 %. This points to the possibility of a partial lithium response with lithium monotherapy and a full (or better) response due to the addition of another mood stabilizer with lithium. The role of psychoeducation and psychological-based therapies (cognitive behaviour therapy, cognitive remediation, interpersonal social rhythms therapy, mindfulness, etc.) might also be an important factor when evaluating the possibility of achieving a better, or a complete, lithium response.

A fifth difficulty that should also be taken in account is the variability of lithium response due to its blood level. As lithium has a narrow therapeutic index, plasma lithium levels can be measured safely and accurately and are reasonable proxies for concentrations in the brain. Titration of plasma levels according to clinical need and symptomatic profile is a useful approach (Kleindienst et al. 2007), and this individualization of lithium dosage, on the basis of polarity, is likely to enhance efficacy and reduce side effects. Severus et al. (2009) has also shown that depression-prone bipolar disorder patients are likely to benefit from prophylactic lithium levels of 0.4–0.8 mmol⁄L, whereas those predisposed to mania tend to benefit from higher levels of 0.6–1.0 mmol⁄L. Thus, as mentioned by Malhi et al. (2011), lithium needs to be measured both literally in terms of its plasma levels and metaphorically, with respect to its clinical and functional effects. For that purpose, Malhi has designed a very useful tool, the ‘lithiumeter’, which provides a visual scale for gauging the optimal lithium plasma level, according to curative (on mania or depression) or prophylactic use.

Finally, a sixth difficulty is probably treatment compliance itself, which plays a key role in the lithium response (Berk et al. 2010). Johnson and McFarland (1996) followed 1,500 patients treated with lithium and reported that the mean duration of continuous lithium adherence was 76 days. Thus, the potential benefits of lithium on recovery, preventing relapse and reducing mortality, are significantly undermined by poor adherence. Although clinicians commonly attribute their patients’ medication non-adherence to side effects, a large survey of more than 3,000 patients suffering from mood disorders found that side effects are not a major determinant of adherence (Morselli and Elgie 2003). The results indicated that only 18.3 % of the patients stated that side effects were the main reason for treatment discontinuation. The major reasons cited for poor adherence were fear of becoming addicted to medication, poor medication information and fear of long-term side effects. This study found that about 35 % of the patients did not receive any information on the possible side effects of their medications and more than 50 % had not received guidance on side effect management. Lack of information may have contributed to the fear of side effects, and this fear may be a stronger predictor of non-adherence than the side effects themselves (Lingam and Scott 2002).

Despite the six levels of difficulties just described, recently published studies define two categories of patients: ‘responders’ and ‘nonresponders’. Given the complexity just discussed, it is not surprising that these definitions may vary from study to study. A classic clinical assessment can be undertaken using the Affective Morbidity Index (AMI) (Coppen et al. 1973) or with the Illness Severity Index (ISI) (Maj et al. 1984). The AMI takes into account the duration and the severity of an episode. Similarly, the ISI was defined as the frequency of affective episodes prior to starting lithium adjusted for age at the time lithium was started. Fifteen years ago, Canadian researchers introduced a scale allowing quantitative retrospective assessment of the quality of prophylactic lithium response (Manchia et al. 2013). This scale is referred to informally as ‘the Alda scale’. Criterion A rates the degree of response (activity of the illness with an adequate lithium treatment) on a ten-point scale. Criterion B weighs clinical factors considered relevant in determining whether the observed clinical change is due to lithium treatment or not. Criterion B involves B1, the number of episodes off the treatment; B2, frequency of episodes off the treatment; B3, the duration of treatment; B4, compliance during period(s) of stability; and B5, the use of additional medications during periods of stability. The total score is obtained by subtracting B from A and is in the range of 0–10. It has been used widely, for example, in the Consortium of Lithium Genetics (ConLiGen) project, which sought to conduct a genome-wide association study (GWAS) in a large population of lithium-treated patients.

In clinical studies, the definition of lithium response has to specify the design and the aim of the trial: the diagnostic group studied (bipolar I or bipolar II), the long-term prophylactic response or the curative effectiveness of episodes, the number (and the name) of mood stabilizers associated with lithium, the association with psychoeducation and/or psychological and social therapies and the frequency of serum levels and their results. Remarkably, an agreed definition for lithium response(s) is yet to be written after more than 60 years of lithium prescription.

First, time to response is a relevant factor that appears highly variable in clinical trials. The global consensus is that, in the first year of treatment, morbidity on lithium may still be high in patients who respond fully in the long term (Ahrens et al. 1993). The difference in time needed to obtain a significant response may reflect the heterogeneity of lithium’s mechanisms of action as well as differences in metabolism between fast and slow responders. This difference could also be linked to compliance issues, time needed to achieve effective yet tolerable blood levels, psychological factors and interactions between the natural course of the illness and other environmental factors (addictions, antidepressant prescriptions, lack of insight, etc.). Consequently, to properly address these issues, prospective studies will be needed, with large samples of bipolar disorder patients treated with lithium.

Most recent clinical drug trials have been relatively short in duration and, therefore, do not accurately assess maintenance or prophylaxis. Moreover, other analyses that have tested lithium response have often relied on variables of convenience available in samples collected for other purposes. Fortunately, there are some exceptions, and we should acknowledge the work of some international groups, such as the International Group for the Study of Lithium-treated patients (IGSLI), which has collected data for decades (Grof 1994, 2006; Schultze et al. 2010).

The methodology on lithium response must be proportionate with the complexity of the issue. To identify the predictors of lithium response, the patient’s profile and treatment adequacy should be carefully assessed, since there are numerous confounding factors—and most variables are highly correlated. For example, Baethge et al. (2003) found that earlier lithium initiation was strongly associated with greater pretreatment illness intensity. This association leads to greater apparent reduction of morbidity during treatment versus before treatment and could theoretically indicate early treatment as a predictor of good response. However, greater morbidity probably encouraged earlier treatment; thus this inference could be misleading.

Trial methodology is crucial to the establishment of valid predictors. Some authors argue for establishing multivariate methods to take this complexity into account (Grof et al. 1993), while others highlight effect-size measure between studies in a meta-analytic approach (Kleindienst et al. 2005a, b). Each method has its own limitations. Multivariate analysis integrates multiple variables in the same model of analysis, and its hypothesis-free nature can potentially lead to identifying new variables associated with lithium response. The meta-analysis approach gives some strength to a candidate variable as a predictor; however it is subject to bias from aggregated data collected from different populations and bias from multiple comparisons.

As a unique definition for lithium response is impossible to find in the studies conducted to date, with variable durations of lithium prescription and with unavoidable methodological bias, one can easily understand why most of the clinical trials dedicated to lithium response may be skewed. The main results will be summarized in the following paragraphs; however one should keep in mind the potential reasons for these frequent contradictory findings.

Some studies have shown that patients who have relatives with bipolar disorder derive more benefit from lithium prophylaxis than do those whose family history is negative (Grof 1994). These early studies were retrospective, and it is important to note that other authors found contrary results (Coryell et al. 2000). Over the past two decades, there has been growing interest in longitudinal studies of the children of parents with bipolar disorders (Duffy et al. 2014). It has been advanced that a good response to lithium in a parent could be a predictor of response in the offspring. Even though this assumption is frequently supported by expert opinion, the available data remain tenuous (Duffy et al. 2007).

The response to long-term lithium may cluster in families (Grof 2006). Only lithium responders have been observed to have significant excesses of bipolar disorders in their family history. The first-degree relatives of bipolar patients responding to lamotrigine have an overabundance of anxiety disorders, panic attacks, substance abuse and alcohol addictions (Grof 2003).

Data concerning first-episode polarity associated with lithium response are sparse. Shapiro (1989) compared bipolar patients treated with lithium, imipramine or both on recurrence rate. He suggested that bipolar patients with a manic index have a better lithium response. However, the study by Yazici et al. (1999), including more than 300 bipolar patients, found a relationship between the index episode being manic and poor lithium response.