Glycaemic control in people with type 2 diabetes mellitus during and after cancer treatment: A systematic review and meta-analysis

Background Cancer and Diabetes Mellitus (DM) are leading causes of death worldwide and the prevalence of both is escalating. People with co-morbid cancer and DM have increased morbidity and premature mortality compared with cancer patients with no DM. The reasons for this are likely to be multifaceted but will include the impact of hypo/hyperglycaemia and diabetes therapies on cancer treatment and disease progression. A useful step toward addressing this disparity in treatment outcomes is to establish the impact of cancer treatment on diabetes control. Aim The aim of this review is to identify and analyse current evidence reporting glycaemic control (HbA1c) during and after cancer treatment. Methods Systematic searches of published quantitative research relating to comorbid cancer and type 2 diabetes mellitus were conducted using databases, including Medline, Embase, PsychINFO, CINAHL and Web of Science (February 2017). Full text publications were eligible for inclusion if they: were quantitative, published in English language, investigated the effects of cancer treatment on glycaemic control, reported HbA1c (%/mmols/mol) and included adult populations with diabetes. Means, standard deviations and sample sizes were extracted from each paper; missing standard deviations were imputed. The completed datasets were analysed using a random effects model. A mixed-effects analysis was undertaken to calculate mean HbA1c (%/mmols/mol) change over three time periods compared to baseline. Results The available literature exploring glycaemic control post-diagnosis was mixed. There was increased risk of poor glycaemic control during this time if studies of surgical treatment for gastric cancer are excluded, with significant differences between baseline and 12 months (p < 0.001) and baseline and 24 months (p = 0.002). Conclusion We found some evidence to support the contention that glycaemic control during and/or after non-surgical cancer treatment is worsened, and the reasons are not well defined in individual studies. Future studies should consider the reasons why this is the case.


Introduction
Incidence of diabetes mellitus (DM) continues to grow worldwide with 415 million estimated cases in 2015 and figures are predicted to reach 642 million by 2040 [1]. Current estimates in the United Kingdom suggest approximately 6% of the population has DM if both diagnosed and undiagnosed cases are included [2]. There is growing evidence that for individuals with DM, the risk of developing cancer significantly increases when compared to a non-diabetic population [3][4][5]. This is particularly the case for liver, pancreatic, colon/rectum, breast and bladder cancers [4]. For example, a threefold increase in the risk of developing colorectal cancer has been described in Type 2 DM (T2DM) populations [6].
Studies investigating consequences of cancer treatment on people with DM report worse outcomes when compared to non-diabetic counterparts. Consequences include: increased mortality [7,8], higher infection rates [9,10], higher hospitalisation rates [10], worse physical function [11] and poorer prognosis [12,9]. Potential reasons for these poorer outcomes include: prioritising cancer treatments over DM self-management activities [11,13]; increased prevalence of and/or under recognition of hyperglycaemia [14]; and clinicians lacking skills in managing both these complex conditions [14,15]. Sub-optimal DM management during cancer care can result in worsened glycaemic control [13]. This has been associated with lower overall survival [11,16] and may be explained, to some extent, by increased deaths from cardiovascular disease [17] and obesity.
Formal guidance on managing patients with comorbid cancer and DM is limited despite a number of epidemiological studies identifying an increased risk of developing cancer in people with T2DM [18,19] and potential interactions between treatments for these two diseases [20]. Likewise little is known about the short-term impact of cancer treatment/care on glycaemic control. One paper found that adherence to glucose lowering drugs decreased in patients following cancer diagnosis [21]. To improve outcomes for patients with both cancer and DM, health professionals need a greater understanding of the impact cancer treatment has on glycaemic control and concomitant DM self-management activities [11,13]. Consequently, this review was undertaken to answer the following question: • Does glycaemic control worsen during treatment for cancer in people with T2DM?

Methods
To answer the questions outlined above, we undertook a systematic review and meta-analysis of published quantitative studies investigating glycaemic control in patients with comorbid cancer and T2DM. We used HbA1c levels as an indication of glycaemic control. The recommended levels should be individually determined according to age, body weight, concomitant complications and diabetes duration. Most adults benefit from HbA1c 53 mmols/mol (7%) and a clinically relevant difference is considered to be change in HbA1c of 6mmols/mol, (0.5%) [22,23].

Search strategy
We developed a broad search strategy (Table 1)  • included people with pancreatic cancer due to the higher proportion of pancreatic cancer patients having diabetes and glucose intolerance [24] • investigated effects of cancer treatment on glycaemic control of Type 1 DM, where type 1 DM share a lesser number of concordant predictors with cancer then T2DM • reported qualitative findings only

Data extraction and quality appraisal
In accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [25], data on outcomes relating to glycaemic control in patients with comorbid cancer and T2DM were extracted systematically. We developed standardised forms, and study data were extracted by SP. Methodological study quality was assessed using the Effective Public Health Practice Project (EPHPP) Quality Assessment Tool for quantitative studies [26]. Studies were independently reviewed by two researchers (SP and JA) and disagreements were resolved through discussion. Studies were judged to be low quality if they had a small sample size, did not describe the study sample, did not describe the study time frame, did not justify the outcome measures used, failed to control for confounding variables or did not fully explain statistical analyses performed. Moderate to high-quality papers met some or all of these criteria. Low-quality papers were not excluded, but results are given less emphasis in the discussion. Tables summarising key findings were created.

Meta-analysis
We were interested in measuring HbA1c (%/mmols/mol) change over time post baseline (diagnosis and/or cancer treatment initiation). The time points of interest were selected based on available data as follows; baseline (T0), one year post-baseline (T1), two years post-baseline (T2).
Of the papers eligible for inclusion the following data were extracted: mean, standard deviation (SD) and number of participants. Missing SDs were imputed using metagear etd software package in R. This uses the coefficient of variation from all complete cases to calculate missing values [27,28]. Datasets with imputed values were then analysed using a random effects model within the metafor software package [29]. This model does not make an inference based on just the studies analysed, but sees them as a random sample of studies from a larger population of studies that could or have been done; thus it allows for estimation of heterogeneity and inference to a broader population of studies.
To test if there was a statistically significant difference between the three time periods the means, SDs and variance estimates were extracted from each of the analyses; and a mixedeffects model analysis undertaken in metafor using the time period (baseline, one year following, two years following) as the moderator variable. The three pooled estimates were themselves entered into a fixed-effects regression model with the time point as the moderating factor. This allowed for calculation of both change over these time periods compared to the baseline, and the amount of heterogeneity explained by the model; as well as confidence intervals and p-values. Full data and code are shown in S1 Appendix.

Results
The literature search generated 6,886 potential papers of which 775 were duplicates. Following title and abstract review, 6,067 were excluded (Fig 1). Of 44 papers read in full, 36 were excluded and eight papers were selected for review (for further details on excluded papers, please refer to Fig 1).
All eight studies employed a retrospective review of case notes and/or medical information design focused on a convenience sample of patients with cancer and pre-existing T2DM. Of these, five studies used a within group comparison using the same participants who had T2DM before and after receiving cancer treatment [30,31,33,35,37]. Three studies compared cancer patients with pre-existing T2DM to cancer patients with no history of T2DM, before and after cancer treatment [32,34,36]. A total of 6,433 people participated in the included studies and sample sizes were diverse, ranging from 29 [33] to 4,474 participants [34]. Participants' mean age ranged from 59.0 years [37] to 75.0 years [33]. Based on diagnostic groups studied; three studies included male-only participants [32][33][34], one study included female-only participants [31] and four studies included both male and female participants [30,35,36,37].
All studies assessed the impact of cancer diagnosis and/or cancer treatment on T2DM outcome, however timing of assessments varied between studies. Whilst for two studies [32,36] time frames were unclear, for the remaining studies the time interval from first to last data collection points ranged from 12 [35,37] to 84 months [30]. Most studies focused on one cancer type, with the exception of one study which included patients with either breast, colon or prostate cancer [30]. Of those studies focusing on one cancer type, three included men with prostate cancer [32][33][34], one included people with myeloma [36], one included women with breast cancer [31] and two studies included people with gastric cancer [35,37].
Measurement of glycaemic control. All selected studies routinely assessed HbA1c (%/mmols/mol) as a long term determinant of glycaemic control. Mean percentage HbA1c level (and mmol/mol) at different time points are described in Table 3 and Table 4.   In the study by Haidar and colleages [33], between five and eight HbA1c levels were recorded throughout a 24 month period (mean HbA1c = 9.3%, 78 mmol/mol).

Quality assessment
Most studies were either of moderate [30,31,35] or weak quality [32,33,36,37] and one was judged to be of strong quality [34]. Detailed evaluation of each study using the EPHPP criteria is shown in Table 5. Generally, papers described as being of weak quality had small sample sizes from a single clinical setting [33], failed to describe if data were missing and/or how this was dealt with [32,33,36,37], excluded cases lacking complete data without sufficiently describing details of the process [33] and did not appropriately describe or deal with confounding variables [32,33,36,37]. Papers described as being of moderate quality failed to provide sufficient detail on the patients' cancer such as type and stage [30,35], details of how patients were selected and why [35], how missing data was dealt with [30], or did not appropriately deal with confounding variables [31].
Glycaemic control. Four studies found no significant difference in HbA1c levels for diabetic patients before and after cancer treatment initiation [30,35] or diabetic patients before and after cancer treatment initiation compared to non-diabetic controls [32,36]. Two studies reported an increase in HbA1c after androgen deprivation therapy (ADT) was initiated [33,34] and one study described raised HbA1c levels following diagnosis and treatment (not specified) for breast cancer [31]. One study in gastric cancer showed mixed results according to type of surgical procedure and/or BMI, however the study design was judged to be weak [37].
Of the four studies measuring FBG, two described a significant increases in FBG at 24 months [33] and 60 months [32] after receiving a course of ADT. However both were evaluated as weak quality studies and reported results for time bands rather than specified time points. For people with gastric cancer one study reported a significant decrease in FBG within 12 months following gastrectomy [35] whilst the other showed reduction only for those with a high BMI who underwent a total gastrectomy and modified Roux-en-y anastomosis and those with normal BMI who underwent a total gastrectomy and standard Roux-en-y anastomosis [37].
In three studies measuring insulin resistance, one study described a significantly lower resistance to insulin post-gastrectomy [35] and two studies reported increased insulin requirements after patients commenced ADT [33,34].
Risks and side effects. Adherence to diabetic medications from the time of cancer diagnosis was described in three studies. For current DM medication users, patients reported they either ceased taking DM medications [35], or reduced DM medications use after their cancer Glycaemic control in people with diabetes mellitus during and after cancer treatment: A systematic review diagnosis [31,35]. The proportion of patients achieving recommended HbA1c levels decreased following worsened adherence to DM medication after being diagnosed with breast cancer [31]. Following diagnosis, one study reported that prostate cancer patients were prescribed new DM medication, or changed their current DM medication to a new class once commencing ADT [34]. Drug classes included metformin, sulfonylureas, other oral drugs and insulins. Women with breast cancer identified as being non-adherent with oral T2DM medication were more likely to be taking !4 cardiovascular disease medications then adherent users [31].

Meta-analysis
Six of the seven papers included in the systematic review were included in the meta-analysis. The paper by Haidar and colleagues [33] was excluded as it only reported median and range, the data was too heterogeneous, the study time points did not match those of interest and no response was received from the corresponding author in attempt to clarify these points. The paper by Bayliss and colleagues [30] was excluded as the data were too heterogeneous and the study time points did not match those of interest.
We compared the mean % (mmol/mol) HbA1c levels and SDs. Although all studies provided mean scores, SDs were missing from some and so we calculated SDs from standard errors where provided [31,34]. For studies which provided means and ranges [32,35,36] SDs were imputed using metagear [38].

Discussion
Results from the meta-analysis demonstrate deterioration in HbA1c levels at 24 months postcancer diagnosis and treatment initiation. When limiting the analysis to exclude studies with populations receiving gastric surgery, HbA1c levels increase at both 12 and 24 months in comparison to baseline.
All studies included routinely assessed HbA1c as a long term determinant of blood glucose levels. Four studies found no significant difference in HbA1c levels before and after cancer treatment [30,32,35,36], three reported an increase in HbA1c levels following cancer diagnosis and treatment [31,33,34] and one reported mixed results depending on participants BMI and type of surgical procedure undergone [37]. This discrepancy may be explained by differences in BMI resulting from the specific cancer diagnosis and/or its corresponding treatment. For example, the first line treatment for gastric cancer is radical surgery which removes some or all of the stomach and limits food intake [35,37]. An alternative explanation for this discrepancy might be that clinically significant changes in glycaemic control necessitated adjustment of diabetes therapy however this was not reflected in changes in HbA1C.
Differences in cancer type and cancer treatment across studies may also explain inconsistencies in the results between the studies. For example, the three studies which highlighted adverse effects on glycaemic control included populations that may have received hormone therapy as a part of their cancer care [31,33,34]. Two studies described a significant increase in FBG at up to 24 [33] and 60 months [32] after receiving a course of ADT, suggesting an increase in insulin resistance. On the other hand, two studies [35,37] reported a significant decrease in FBG within 12 months following gastrectomy further suggesting that cancer type and corresponding treatment will have differing effects on glycaemic control. Chemotherapy drug regimens vary according to cancer type/stage however none of the studies provided information on this. Thus it is possible that differences in glycaemic control arise from different treatment regimens being compared. Likewise many patients are administered steroids as part of an anti-emetic regimen however due to the lack of information provided by the included studies it is not possible to assess the impact of this on glycaemic control. Further research is required to investigate this in more detail.
Considerable differences between studies may explain the lack of consistency in the association between HbA1c levels and initiation of cancer treatment. For example, the timing of HbA1c recordings relative to the timing of their cancer diagnosis varied greatly across the studies. The study by Chou and colleagues [36], which was rated as being of weak quality, only measured HbA1c at baseline, therefore there was no comparison. Whilst Derweesh and colleagues [32], also judged as being of weak quality, assessed HbA1c on more than one occasion (baseline and 60 months post-diagnosis or treatment initiation), it is possible patients had either fully recovered by the second assessment or had sufficient time to adapt to these two concurrent illnesses. Generally there was variation between studies in relation to: cancer diagnosis (type, stage, severity), cancer treatment (chemotherapy, endocrine therapy, radiotherapy, surgery, ADT), competing comorbidities (obesity and cardiovascular disease), attention to potential confounders (duration of DM, treatment history, supportive medications) and inconsistencies in inclusion and exclusion criteria, dealing with missing data and use of a control group for comparison.
Poor outcomes such as hyperglycaemia have been reported in patients with comorbid cancer and diabetes [14] but it should be noted that there is little available data to date. For patients with comorbid cancer and diabetes, the adverse outcomes highlighted in this review include higher mortality rates [36], consistent with findings in several studies [7,8,11,16] and poor adherence to DM medications [31,34,35] as reported by Hershey and colleagues [11].
Poorer adherence to diabetic medications post-cancer diagnosis was described in three studies and patients reported they either ceased taking DM medications [35], or reduced DM medications use after diagnosis [31,35]; and this supports findings from the study by Zanders and colleagues [21]. An explanation for this may be that little attention is paid to glycaemic control by cancer health professionals and/or poorer self-management by the patients themselves when also burdened with the added responsibilities and strains associated with other competing chronic conditions [30] including cancer self-management [13], however evidence to support this is extremely limited. It is also possible that there is a lack of integrated care and competing care priorities but again, evidence to support this is limited. It would be useful to know how this group of people assess and decide between competing care priorities. Being able to differentiate between modifiable and non-modifiable factors at both the cancer and T2DM level may help health professionals to identify how best to support and intervene with this group of people.

Limitations and recommendations
The findings of this meta-analysis are limited by the small number of studies reporting on the impact of cancer treatment on glycaemic control and adverse outcome in people with T2DM. This is further compounded by methodological issues and inconsistencies identified in the seven included studies. All used a retrospective design incorporating routinely collected data which varied in terms of data completeness and also time points at which data collection occurred relative to cancer treatment. As the studies were observational there was no suitable comparison group and it is possible that glycaemic control may worsen in participants' without a cancer diagnosis over the 12 to 24 month period [39]. Generally it is not clear whether studies included all patients who passed through the system, or excluded those with missing information. Inconsistencies in key clinical variables made comparisons between studies difficult. A prospective study design may have better resolved some of these issues. Finally we initially intended to evaluate whether DM complications are greater in people with T2DM who receive cancer treatment, however this was not possible due to the limited evidence available.

Concluding remarks and implications for research
This meta-analysis found that following treatment for cancer (and particularly ADT) there is a small statistically significant increase in HbA1c in people with pre-existing T2D when gastric surgery cases are removed. Whilst this increase was not clinically meaningful, results should be treated with caution due the lack of high quality evidence available for review. The limited research conducted evaluating glycaemic control in cancer patients with T2DM means it is not possible to judge potentially serious outcomes as a result of treatment interactions in this population. Understanding such potential interactions and outcomes could help inform decisions made by health care professionals regarding treatments and care pathways. Future research is required to investigate glycaemic control during cancer treatment and what happens to HbA1c levels and DM complications during steroids and cytotoxic and antiemetic regimes. The research may help inform best care for patients with comorbid cancer and DM.