Challenges determining the best target duration of deep molecular response after which to attempt achieving therapy-free remission in chronic myeloid leukaemia

Mathematical models of CML and TFR are often built on assumptions about haematopoiesis that may or may not be valid. Some mathematical models assume leukaemia stem cells are insensitive to TKI-therapy. However, most models consider that the likelihood of achieving TFR is determined by: (1) whether and how fast leukaemia stem cells are eradicated by TKI-therapy; (2) how effectively re-growth of leukaemia stem cells is suppressed by TKI-therapy and/or (3) the proportion of haematopoietic stem cells with BCR::ABL1 at diagnosis [4,5,6,7,8].

After starting TKI-therapy log₁₀-values of results of polymerase chain reaction (PCR)-based measurable residual disease (MRD)-tests targeting BCR::ABL1 decline at an expected slope of α = –0.04 to –0.05 per day followed by a 2^nd much slower decline with an expected slope of β = –0.002 to –0.006 per day (Fig. 1A) [7]. Based on mathematical modelling of clonal expansion and differentiation of leukaemia cells it is often assumed the initial steeper slope α represents the depletion rate of terminally-differentiated leukaemia cells whereas the 2^nd slope β, the decline of leukaemia progenitor and stem cells [4]. The magnitude of β is typically so small that it has been estimated it could take as many as 30 years for TKI-therapy to reduce numbers of leukaemia stem cells to very few or none [9].

A Wide variations of the bi-phasic decline of BCR::ABL1 after TKI-therapy. Displayed curves are adapted from previously published examples [7]. B Increased CML incidence in the A-bomb survivors in the decades after radiation exposure [26]. C Given a target duration of DMR, there are two types of failure: (1) loss of DMR before reaching the target duration [34]; (2) failing TFR after reaching the target duration and then stopping TKI-therapy [18, 33]. (‘prob’ stands for ‘probability’.) D Once median durations of DMR surpass 30 months, cohorts with longer median durations of DMR have only slightly lower TFR failure rates compared to those with shorter median durations of DMR (Table 1) [13,14,15,16,17,18,19,20,21,22]. Slopes were calculated by least-squares regression with each data point having the same weight. (‘mo’ stands for ‘months’.)

Full size image

There are several important considerations when interpreting α and β slopes (Fig. 1A). 1^st, because leukaemia stem cells can cause recurrence, it is often assumed β slope is more predictive of long-term prognosis compared to α slope, which is often predictive of deep molecular response but not TFR success [4, 7, 10, 11]. However, there are contradictory data [12]. 2^nd, the slopes α and β are not correlated (r = –0.1 to +0.2); namely, rapid early response to TKI-therapy does not necessarily translate to more effective eradication of cells able to cause leukaemia recurrence [9, 10]. 3^rd, to confidently estimate a bi-phasic model and slope β using time-series data of BCR::ABL1-testing we often need an extended period (≥1.5 or even ≥2 years) of quantitative monitoring of BCR::ABL1 transcripts during the β phase [7]. 4^th, because the β phase is often after reaching DMR, detection of residual leukaemia cells might be below the limit of detection (LoD) of the quantitative reverse transcription or DNA-based digital droplet PCR assays (i.e. rendering zeros in test results), complicating the estimation of β [7]. The bottom line is that the β slope has not been used in contemporary algorithms for treating CML.

Even if we could precisely quantify numbers of residual leukaemia stem cells, there is considerable variation in the interval between having ≥1 residual leukaemia stem cell and leukaemia recurrence. After discontinuation of TKI-therapy about 38% (range, 28–57%) of persons have molecular recurrence within 6 months, with more recurrences thereafter albeit at a slower rate (Table 1) [13,14,15,16,17,18,19,20,21,22,23,24]. Because different people have different numbers of residual leukaemia stem cells at TKI-therapy-discontinuation it is unknown whether variation in interval to leukaemia recurrence is attributable to variations in numbers of residual leukaemia stem cells, biological differences in individual leukaemia stem cells’ ability to cause recurrence or stochastic events. Namely, not all or any remaining leukaemia stem cells may divide within a person’s remaining lifetime or observation interval.

Data from the Japanese survivors of A-bombs who had blood analyses biennially beginning in 1950s are informative about leukaemia latency post-exposure [25]. There were 33 excess cases of CML 1950–2001 attributable to exposure to the A-bomb radiations [26]. Analyses of paraffin-embedded spleen and bone marrow samples confirmed presence of BCR::ABL1, indicating a similar aetiology to spontaneous cases of CML [27]. 16 excess cases were diagnosed within 10 years of exposure and 17, >10 years after exposure (Fig. 1B) [26]. These data imply a wide range of latency from one leukaemia stem cell to diagnosis of CML. This is especially so in females whose excess CML occurrence rate is significantly higher compared with males >20 years after radiation exposure [28]. (At the other extreme CML can be diagnosed <6 months after birth implying a brief latency [29, 30].)

The premise that substantial depletion of leukaemia stem cells is a pre-requisite for long-term TFR is an unproven assumption. After a few years of sustained DMR a person may have >10,000 residual leukaemia stem cells [9]. Even if he/she is able to maintain major molecular response (MMR; i.e. MR³) >4 years after TKI-discontinuation, with high probability CD26-positive leukaemia stem cells are detectable in his/her blood, although most or all leukaemia stem cells might be quiescent [31, 32]. Such a person could be operationally cured, relapse at a later time or die of something else before CML recurrence. Only time would tell.

Data from the EURO-SKI study suggest the likelihood of achieving TFR is higher with a longer duration of DMR before TKI-therapy-discontinuation [18, 33]. Probabilities of molecular recurrence in <6 months were 56, 53, 50, 47, 44 and 41% with DMR durations of 1, 2, 3, 4, 5 and 6 years, i.e. a difference of 15 percentage points between a DMR duration of 1 and 6 years (equivalent to a decrease of 3 percentage points per additional year of DMR).

However, this reading of the EURO-SKI data fails to consider that when someone in DMR continues TKI-therapy there is a chance he/she might lose DMR thereby becoming ineligible to attempt TFR under current clinical practice guidelines. For example, when persons having maintained 1 year of DMR continue taking TKIs, 1.3, 3, 4.7 and 8.7% lose DMR despite 1, 2, 3 and 5 additional therapy years [34].

The proper calculus to account for everyone with CML follows. Assuming conclusions from the EURO-SKI study apply to everyone with CML and starting with 1000 persons all of whom already have 1 year of sustained MR⁴, if we stop TKI-therapy now 560 persons (= 1000 × 0.56) would have molecular recurrence in <6 months. If, on the other hand, TKI-therapy is continued because we attempt to achieve a target DMR duration of 6 years, 87 (= 1000 × 0.087) would lose MR⁴ before the end of the 6^th year. (It should be emphasized that MR⁴ can be lost and regained, especially in those remaining in MMR [14, 17, 21]. Therefore, TFR might be delayed, but not prevented, in at least some of these 87 persons. Our calculation represents a worst-case scenario.) If the 913 persons remaining in DMR then stop TKI-therapy, 374 (= 913 × 0.41) would relapse in <6 months. Therefore, by attempting to achieve 5 additional years of DMR the total number of persons who become ineligible for attempting TFR or would have a recurrence in <6 months after attempting TFR drops from 56% to 46% (= (87 + 374) / 1000) in contrast to the 41% suggested by an incorrect reading of the EURO-SKI data (Fig. 1C). In other words, the net benefit of attempting to achieve 6-year DMR is two-thirds (= (56 – 46) / (56 – 41)) of that presented in the EURO-SKI dataset. Note that in our calculations all 1000 subjects begin after achieving 1 year of DMR. Conversely, in the EURO-SKI study, subjects started after varying durations of DMR and those who had already become ineligible for attempting TFR were not included in the sample. Data from the EURO-SKI study cannot be used to determine the best target duration of DMR.

In the above calculations we assumed TFR failure rates based on the EURO-SKI study can be applied to all real-world settings, i.e. rate of failed TFR always decreases as duration of DMR increases. However, this assumption is questionable when we examine the scope of published data. We considered 10 studies covering 1689 subjects diagnosed during 2006–2015 receiving imatinib, nilotinib and/or dasatinib. These data indicate the TFR failure rate declines very slowly (–0.05 percentage point per month) once median DMR duration is >30 months (P = 0.90; Table 1; Fig. 1 D) [13,14,15,16,17,18,19,20,21,22]. The studies reviewed in Table 1 are from different time intervals and in different ethnicities and median DMR durations might not accurately describe the heterogeneous distributions of DMR durations in these studies. However, these data suggest the need to reëvaluate the benefit of attempting to achieve a DMR duration >3 years to increase success of achieving TFR. Even if someone fails their 1^st attempt to achieve TFR, with probabilities of 100% (range, 99–100%) and 96% (range, 88–100%) they would regain MMR and DMR, respectively, after re-starting TKI-therapy [14, 17, 21, 35]. Furthermore, if that person attempts to achieve TFR again, with a probability of 48–68% they would be able to maintain MMR for ≥1 year [36,37,38]; the probability of a 2^nd TFR-failure is not correlated with the duration of DMR before the 1^st TFR attempt [36, 37].

Compliance with periodic monitoring of BCR::ABL1 is important to determine real duration of continuous DMR. For example, if people in DMR have BCR::ABL1-testing scheduled every 6 months but there is 50% non-compliance the real duration of DMR is uncertain in many.

BCR::ABL1 transcript type might also influence likelihood of TFR success. Data suggest e14a2 transcript is associated with deeper molecular response and higher TFR success rate [12, 39,40,41].

Another consideration is whether factors extrinsic to leukaemia stem cells such as the immune system and/or microenvironment might correlate with the likelihood of success of maintaining TFR after stopping TKI-therapy [13, 42,43,44,45,46]. This is consistent with the negative correlation between chronic graft-versus-host disease and cumulative incidence of relapse in people with chronic phase CML receiving an allotransplant [47]. However, recipients of solid-organ transplants receiving intensive posttransplant immune-suppression and people with human immune deficiency virus (HIV)-infection do not have higher incidence of CML (in contrast to other cancers such as non-Hodgkin lymphomas) [48,49,50].

When deciding whether to stop or continue TKI-therapy in someone with a DMR of a specified duration the physician’s and the patient’s tolerance levels for uncertainty are sometimes discordant. Physicians who are more risk-averse might postpone stopping [51]. Patient volume, practice setting and years and heuristics such as recall bias influence their recommendations [52,53,54]. Time horizon also matters. For instance, a 35-year-old with moderate therapy-related adverse events (AEs) on TKI-therapy and unable to afford the drug might be comfortable stopping TKI-therapy with only a 35 percent chance of TFR whereas a 70-year-old with substantial co-morbidities but tolerating TKI-therapy well without fiscal constraints might be reluctant to stop TKI-therapy even with a 60–70 percent chance of success.

Serious TKI-related AEs are reported in up to 20 percent of people and they often require dose-reduction or discontinuation or switching to a different TKI [55,56,57,58,59,60,61]. AEs are common reasons for poor compliance of TKI-therapy and to achieve MMR an adherence rate (i.e. numbers of tablets taken divided by numbers prescribed) of >90% is required [62]. In some persons with CML decreasing TKI-dose is an alternative to stopping or switching TKIs [35, 63,64,65]. In some patients TKI dose-reduction does not decrease the likelihood of achieving TFR [66]. However, in others decreasing the TKI-dose might decrease the likelihood of achieving MMR [56]. People >60 years are more likely to discontinue TKI-therapy or require dose-reduction because of intolerance compared with younger persons [57, 60].

It is likely that every patient’s optimal duration of DMR is different, reflecting multiple influencing factors [67].

Identifying the best target duration of DMR after which to attempt TFR would be easier if one could accurately predict the likelihood of success versus risk of losing DMR whilst continuing TKI-therapy. However, even if this were possible we would still have to account for inter-physician and -patient variations in uncertainty tolerance, finances and other extrinsic factors such as anti-leukaemia immunity. Consequently, it is impossible there can be a universal best target duration for DMR. In Monty Python and the Holy Grail Sir Lancelot might have said his favourite colour was pink and still made it across the bridge.