Ethnic disparities in initiation and intensification of diabetes treatment in adults with type 2 diabetes in the UK, 1990-2017: A cohort study.

BACKGROUND: Type 2 diabetes mellitus (T2DM) disproportionately affects individuals of nonwhite ethnic origin. Timely and appropriate initiation and intensification of glucose-lowering therapy is key to reducing the risk of major vascular outcomes. Given that ethnic inequalities in outcomes may stem from differences in therapeutic management, the aim of this study was to identify ethnic differences in the timeliness of initiation and intensification of glucose-lowering therapy in individuals newly diagnosed with T2DM in the United Kingdom. METHODS AND
FINDINGS: An observational cohort study using the Clinical Practice Research Datalink was conducted using 162,238 adults aged 18 and over diagnosed with T2DM between 1990 and 2017 (mean age 62.7 years, 55.2% male); 93% were of white ethnicity (n = 150,754), 5% were South Asian (n = 8,139), and 2.1% were black (n = 3,345). Ethnic differences in time to initiation and intensification of diabetes treatment were estimated at three time points (initiation of noninsulin monotherapy, intensification to noninsulin combination therapy, and intensification to insulin therapy) using multivariable Cox proportional hazards regression adjusted for factors a priori hypothesised to be associated with initiation and intensification: age, sex, deprivation, glycated haemoglobin (HbA1c), body mass index (BMI), smoking status, comorbidities, consultations, medications, calendar year, and clustering by practice. Odds of experiencing therapeutic inertia (failure to intensify treatment within 12 months of HbA1c >7.5% [58 mmol/mol]), were estimated using multivariable logistic regression adjusted for the same hypothesised confounders. Noninsulin monotherapy was initiated earlier in South Asian and black groups (South Asian HR 1.21, 95% CI 1.08-1.36, p < 0.001; black HR 1.29, 95% CI 1.05-1.59, p = 0.017). Correspondingly, no ethnic differences in therapeutic inertia were evident at initiation. Intensification with noninsulin combination therapy was slower in both nonwhite ethnic groups relative to white (South Asian HR 0.80, 95% CI 0.74-0.87, p < 0.001; black HR 0.79, 95% CI 0.70-0.90, p < 0.001); treatment inertia at this stage was greater in nonwhite groups relative to white (South Asian odds ratio [OR] 1.45, 95% CI 1.23-1.70, p < 0.001; black OR 1.43, 95% CI 1.09-1.87, p = 0.010). Intensification to insulin therapy was slower again for black groups relative to white groups (South Asian HR 0.49, 95% CI 0.41-0.58, p < 0.001; black HR 0.69, 95% CI 0.53-0.89, p = 0.012); correspondingly, treatment inertia was significantly higher in nonwhite groups at this stage relative to white groups (South Asian OR 2.68, 95% CI 1.89-3.80 p < 0.001; black OR 1.82, 95% CI 1.13-2.79, p = 0.013). At both stages of treatment intensification, nonwhite groups had fewer HbA1c measurements than white groups. Limitations included variable quality and completeness of routinely recorded data and a lack of information on medication adherence.
CONCLUSIONS: In this large UK cohort, we found persuasive evidence that South Asian and black groups intensified to noninsulin combination therapy and insulin therapy more slowly than white groups and experienced greater therapeutic inertia following identification of uncontrolled HbA1c. Reasons for delays are multifactorial and may, in part, be related to poorer long-term monitoring of risk factors in nonwhite groups. Initiatives to improve timely and appropriate intensification of diabetes treatment are key to reducing disparities in downstream vascular outcomes in these populations.

Introduction

The burden of type 2 diabetes mellitus (T2DM) is growing worldwide, disproportionately affecting individuals of nonwhite ethnic origin [1-3]. Around 6% of the UK population have T2DM, with the risk of developing T2DM approximately 2- to 4-fold greater in migrant populations of South Asian or black African or Caribbean descent. Additionally, once diagnosed with T2DM, the lifetime risk of developing cardiovascular conditions is higher again in minority ethnic groups than those of white European origin [4-11].

Timely and appropriate initiation and intensification of glucose-lowering therapy have been shown to substantively reduce adverse diabetes outcomes [12,13]. In the UK, clinical guidelines for the therapeutic management of blood glucose recommend that newly diagnosed individuals initiate with noninsulin monotherapy, usually metformin, if their glycated haemoglobin (HbA1c) level is above 6.5% (48 mmol/mol). Intensification with additional noninsulin therapies, and ultimately insulin, is recommended if HbA1c remains above 7.5% (58 mmol/mol) [14].

Poor glycaemic control is associated with an increased risk of macrovascular complications (such as coronary disease and stroke) and microvascular complications (such as diabetic retinopathy, neuropathy, and kidney disease). One key driver of suboptimal glucose management is therapeutic inertia, defined as failure to appropriately initiate or intensify treatment in a timely manner following identification of uncontrolled risk factor levels. In the UK, it has been shown that one-third of individuals do not achieve target HbA1c levels of 7.5% (58 mmol/mol), despite clinical guidelines recommending that therapy be intensified within 3–6 months of identifying a raised HbA1c [14,15]. Furthermore, a 2019 UK study has highlighted the high economic burden associated with therapeutic inertia, with costs related to increases in diabetes-related complications and lost workplace productivity equalling £2.6 million over the next 10 years [16].

Given that ethnic inequalities in diabetes care and outcomes can accumulate from even before diagnosis, it is essential to identify where along the care pathway disparities arise. In particular, it is essential to identify easily modifiable aspects of the pathway that may contribute to inequalities. As such, the aim of this study is to identify ethnic differences in the timeliness of initiation and intensification of glucose-lowering treatment in individuals with newly diagnosed T2DM in the UK—specifically, (1) ethnic differences in the ordering of glucose-lowering treatments, (2) ethnic differences in time to initiation and intensification of treatment, and (3) ethnic differences in treatment delays (therapeutic inertia) following identification of uncontrolled HbA1c.

Methods

Study design and population

An observational cohort study utilising the Clinical Practice Research Datalink (CPRD) was undertaken. The CPRD is a clinical research database containing anonymised longitudinal primary care records for approximately 15 million people from 714 general practices and has been shown to be representative of the UK population with respect to age, sex, and ethnicity [17]. Adults aged 18 and over who were registered between January 1990 and December 2017, with at least 12 months of continuous registration prior to first recorded diagnosis of T2DM, were included in the study. T2DM was identified using an adjudication algorithm developed to minimise misclassification of diabetes status and type in electronic health records (EHRs) [18].

Transitions between treatment stages were analysed in three time periods: (1) time from diagnosis to initiation of noninsulin monotherapy, (2) time from noninsulin monotherapy to intensification with noninsulin combination therapy, and (3) time from noninsulin combination therapy to intensification with insulin therapy.

Covariates

Self-reported ethnicity was identified using Read codes and collapsed into the five main categories of the 2001 UK census (white, South Asian, black African/Caribbean, mixed, and other). For individuals with more than one ethnicity code on their primary care record, an algorithm was used to assign a best ‘single’ ethnicity—based on the most commonly and most recently recorded codes (S1 Fig) [19]. Age at diagnosis was calculated by subtracting the year of birth from year of diagnosis and was divided into 10-year age bands. Deprivation was measured using quintiles of the Index of Multiple Deprivation (IMD) [20].

For the time from diagnosis to initiation with noninsulin monotherapy, baseline was defined as the date of T2DM diagnosis. Baseline covariates were identified from the value closest to the date of T2DM diagnosis from the 12 months preceding or the 3 months following diagnosis. These included HbA1c, body mass index (BMI), systolic blood pressure (SBP) and diastolic blood pressure (DBP), and smoking status (categorised as ‘never smoker’, ‘current smoker’, and ‘ex-smoker’). Although BMI was considered as a continuous covariate in the analytical models, descriptive statistics categorised BMI according to the standard categories of underweight, normal weight, overweight, and obese, with adjustments for South Asian ethnicity as recommended by WHO, which classify normal weight as between 18.5 and 22.9 kg/m2, overweight as between 23 and 27.4 kg/m2, and obesity as 27.5 kg/m2 or over for this ethnic group [21].

Comorbidities were considered present at baseline if recorded at any time prior to diagnosis. These included depression, macrovascular disease (hypertension, myocardial infarction, angina, stroke, and heart failure), and microvascular disease (chronic kidney disease, retinopathy, and neuropathy) (see S1 Table for all code lists). The number of consultations and oral medications in the 6 months prior to diagnosis was also included as baseline covariates.

For the time from noninsulin monotherapy to noninsulin combination therapy, baseline was defined as the date of initiation of noninsulin therapy. Baseline HbA1c, BMI, and smoking status were derived from the date closest to the date of monotherapy initiation in the 6 months preceding. Counts of consultations and oral medications were calculated from the 6 months prior to monotherapy initiation. Comorbidities were considered present if recorded at any time prior to monotherapy initiation. For the time from noninsulin combination therapy to insulin therapy, baseline was defined as the date of initiation of noninsulin combination therapy, with baseline HbA1c, BMI, smoking status, counts of consultations and oral medications, and comorbidities defined as above.

For each of the three time periods of interest, additional between-treatment variables were constructed. These included the number of HbA1c measurements and the number of consultations between treatment stages.

Glucose-lowering treatment

Glucose-lowering treatment was identified from GP prescribing data and categorised into six classes: metformin, sulfonylurea, newer agents (dipeptidyl peptidase-4 inhibitor [DPP4i], sodium-glucose cotransporter-2 inhibitor [SGLT2i], and glucagon-like peptide-1 [GLP-1] receptor agonists), thiazolidinediones, insulin, and other drugs. Newer agents were grouped together because they represent treatment strategies used for similar stages of diabetes progression.

Statistical analysis

Patterns of glucose-lowering treatment

Firstly, the proportion of individuals prescribed each class of glucose-lowering drug was calculated from the initiation of a single drug through to the addition of up to four more drugs and compared between ethnic groups. Secondly, patterns of treatment up-titration were compared between ethnic groups using sequence analysis. Sequences included (1) staying on a single drug regime for the entire follow-up period, (2) moving between two drug regimes, and (3) moving between three drug regimes.

Time to treatment initiation and intensification

Multilevel multivariable proportional hazard models assuming an exponential baseline hazard function were employed to identify ethnic differences in time to initiation and intensification of glucose-lowering treatment while accounting for the clustering of individuals within general practices. Individuals free from glucose-lowering treatment at the date of T2DM diagnosis were eligible for inclusion. Individuals who initiated diabetes treatment in the 90 days prior to diagnosis were considered to be ‘baseline initiators’ and, for analytical purposes, had their initiation date moved to 1 day after diagnosis to allow entry into the cohort. For initiation of noninsulin monotherapy, follow-up time began at the date of T2DM diagnosis and ended at the date of initiation. For intensification to noninsulin combination therapy, follow-up time began at commencement of monotherapy and ended at the date of intensification with combination therapy. For intensification to insulin therapy, follow-up time began at the commencement of noninsulin combination therapy and ended at the date of intensification with insulin.

For individuals who did not initiate or intensify with the drug of interest during the follow-up period, follow-up time was censored at the earliest of the following: date of intensifying to a different drug, death, transferring out of the practice, or last data collection. For example, individuals who initiated with a treatment other than noninsulin monotherapy were censored at the date the alternative treatment was commenced, and individuals who intensified directly from noninsulin monotherapy to insulin were censored at the date of insulin initiation.

All models adjusted for hypothesised covariates as identified in the directed acyclic diagram (S2 Fig); namely, age at diagnosis, sex, deprivation, year of diabetes diagnosis (to account for secular trends in prescribing guidelines and treatment availability), HbA1c, BMI, and smoking status at the start of each follow-up period; presence of depression, micro- and macrovascular comorbidities at the start of each follow-up period; and count of consultations and oral medications in the 6 months prior to the start of each follow-up period. Models for intensification to combination therapy and insulin therapy additionally adjusted for time since diagnosis. As individuals attending the same general practice may have similar levels of care provision and clinical coding, multilevel modelling was used to account for the clustering of people within practices.

Because of overdispersion in the Poisson model, multilevel multivariable negative binomial regression was used to estimate ethnic differences in the count of HbA1c measurements, and consultations between treatment stages adjusted for all hypothesised confounders and clustering by practice.

Ethnic differences in therapeutic inertia

As there is no single accepted definition of therapeutic inertia, we adopted a definition used in previous UK database studies [22-24]. For the purposes of our study, therapeutic inertia was defined as the failure to intensify treatment within 12 months of having an HbA1c of >7.5% (58 mmol/mol) recorded by the general practitioner. Individuals with any raised HbA1c following (1) diagnosis, (2) initiation of noninsulin monotherapy, or (3) initiation of noninsulin combination therapy with at least 12 months of follow-up were included in the analysis. Models to estimate ethnic differences in the odds of experiencing therapeutic inertia were constructed adjusting for all hypothesised confounders. Models for treatment inertia around time of intensification to combination therapy and insulin therapy additionally adjusted for time between first raised HbA1c and T2DM diagnosis in each of the three time periods. Descriptive statistics for factors associated with therapeutic inertia including the number of HbA1c’s above 7.5% (58 mmol/mol) measured between first raised HbA1c and treatment intensification, the number of consultations between first raised HbA1c and treatment intensification, and the proportion of individuals with a consultation within 3 months of the first recorded raised HbA1c were calculated and presented by ethnic group.

For all analyses, individuals of white ethnicity were considered as the reference population. Comparisons between the white, South Asian, and black African/Caribbean populations are reported in the main results. Analyses were conducted as specified in the scientific protocol (see S1 Text). Due to the fact that routinely recorded data in primary care EHRs are likely to be missing not at random, multiple imputation of missing data was not considered appropriate because the assumption of missing at random was unlikely to be met. As such, a complete case analysis approach was employed [25]. All analyses were completed using Stata version 15 and reported according to the RECORD guidelines (see S1 RECORD Checklist).

Secondary analysis

Firstly, as current guidelines for diabetes management in the UK suggest a threshold of 6.5% (48 mmol/mol) for initiation of diabetes treatment, a secondary analysis of ethnic differences in treatment inertia at initiation of noninsulin monotherapy was conducted using >6.5% at the definition of raised HbA1c instead of >7.5% (58 mmol/mol). Secondly, as 12 months’ minimum follow-up was required for inclusion into the analysis of treatment inertia, the ethnic breakdown of those included in the analysis was compared to that of those excluded for lack of 12 months’ follow-up at each stage.

Ethical approval

Ethical and scientific approval for this study were granted by the Independent Scientific Advisory Committee (ISAC) (protocol 17_087R) and the London School of Hygiene & Tropical Medicine (project ID 13409).

Results

From a total of 425,811 adults aged 18 and over who were diagnosed with T2DM between 1990 and 2017, 162,238 individuals of white, South Asian, or black ethnicity with at least 1 year of continuous registration prior to diagnosis were included in the study (Fig 1). Individuals of mixed/other ethnicity (n = 2,794) and unknown ethnicity (n = 75,258) were excluded from the study population (comparisons between the white, other, and unknown ethnic groups are reported in S2–S4 Tables). Individuals in the study cohort contributed a mean of 6.2 years of follow-up. The ethnic breakdown of the study cohort was 92.9% white (n = 150,754), 5.0% South Asian (n = 8,139), and 2.1% black (n = 3,345). Compared with the white group, age at diagnosis was markedly younger in ethnic minority groups (South Asian, 55 years; black, 56 years; white, 63 years), and baseline HbA1c was higher (South Asian, 65.8 mmol/mol; black, 69.3 mmol/mol; white, 64 mmol/mol) (Table 1).

Fig 1

Population inclusion flowchart.

CPRD, Clinical Practice Research Datalink; T2DM, type 2 diabetes mellitus.

Table 1

Baseline characteristics stratified by ethnic group.

Baseline characteristics	White		SA		Black
N	150,754		8,139		3,345
Years of follow-up (mean, SD)	6.3	(4.6)	5.9	(4.7)	5.1	(4.4)
Age at diagnosis (mean, SD)	63.4	(13.2)	53	(12.9)	55.8	(13)
Gender, male, n (%)	82,619	(54.8)	4,411	(54.2)	1,679	(50.2)
Deprivation quintile
1 (least deprived), n (%)	27,606	(18.3)	1,030	(12.7)	191	(5.7)
2, n (%)	30,092	(20)	1,321	(16.2)	306	(9.1)
3, n (%)	33,318	(22.1)	1,721	(21.1)	717	(21.4)
4, n (%)	28,989	(19.2)	1,803	(22.2)	929	(27.8)
5 (most deprived), n (%)	30,749	(20.4)	2,264	(27.8)	1,202	(35.9)
Smoking status
Never smoker, n (%)	51,565	(34.2)	4,619	(56.8)	1,721	(51.4)
Current smoker, n (%)	23,503	(15.6)	843	(10.4)	342	(10.2)
Ex-smoker, n (%)	47,629	(31.6)	828	(10.2)	488	(14.6)
Missing, n (%)	28,057	(18.6)	1,849	(22.7)	794	(23.7)
BMI
BMI at diagnosis, kg/m2 (mean, SD)	31.7	(6.1)	29.2	(5.2)	31.3	(5.9)
Underweight (<20, <18.5 for SA)	1,447	(1)	23	(.3)	24	(.7)
Normal weight (20–25, 18.4–23 for SA)	13,560	(9)	532	(6.6)	309	(9.2)
Overweight (25–30, 23.5–27.5 for SA)	39,826	(26.4)	2,251	(27.8)	917	(27.4)
Obese (>30, >27.5 for SA)	71,664	(47.5)	4,013	(49.5)	1,521	(45.5)
Missing	24,257	(16.1)	1,287	(15.9)	574	(17.2)
HbA1c
HbA1c at diagnosis % (mean, SD)	8	(2.1)	8.2	(2.1)	8.5	(2.4)
HbA1c at diagnosis, IFCC (mean, SD)	63.6	(23.2)	65.8	(22.7)	69.3	(26.7)
≤7.5%, n (%)	64,642	(42.9)	3,285	(40.4)	1,284	(38.4)
7.5%–7.9%, n (%)	11,191	(7.4)	767	(9.4)	305	(9.1)
8.0%–8.9%, n (%)	13,291	(8.8)	847	(10.4)	327	(9.8)
≥9.0%, n (%)	29,594	(19.6)	1,692	(20.8)	849	(25.4)
Missing, n (%)	32,036	(21.3)	1,548	(19)	580	(17.3)
Blood pressure
SBP at diagnosis (mean, SD)	140.3	(18.8)	133.2	(17.6)	137.5	(18.5)
DBP at diagnosis (mean, SD)	80.9	(10.9)	80.8	(10.5)	82.5	(10.8)
<140/90, n (%)	27,185	(19)	1,131	(14.8)	671	(21.1)
<150/90, n (%)	20,424	(14.3)	737	(9.6)	471	(14.8)
<        130/80, n (%)	76,154	(53.2)	3,384	(44.3)	1,683	(52.9)
Missing, n (%)	7,564	(5)	500	(6.1)	166	(5)
Comorbidities and medications
Any macrovascular, n (%)	21,669	(14.4)	685	(8.4)	198	(5.9)
Any microvascular, n (%)	4,685	(3.1)	202	(2.5)	96	(2.9)
Depression, n (%)	34,246	(22.7)	1973	(24.2)	721	(21.6)
On antihypertensive at diagnosis, n (%)	43,427	(28.8)	1,428	(17.5)	555	(16.6)
On statin at diagnosis, n (%)	77,072	(51.1)	3,641	(44.7)	1,295	(38.7)
DM treatment initiation characteristics
Initiate <1 year before diagnosis, n (%)	10,654	(7.1)	691	(8.5)	282	(8.4)
Initiate in 12 months prior to diagnosis, n (%)	33,034	(21.9)	2,347	(28.8)	1,086	(32.5)
Initiate within 90 days of diagnosis, n (%)	27,804	(18.4)	1,609	(19.8)	704	(21)
Initiate >90 days after diagnosis, n (%)	45,187	(30)	2,293	(28.2)	666	(19.9)
Noninitiators of DM treatment, n (%)	34,075	(22.6)	1,199	(14.7)	607	(18.1)

Baseline measures of HbA1c, BP, and CVD risk defined as value closest to diagnosis date in the 12 months prior or 3 months after.

Abbreviations: BMI, body mass index; BP, blood pressure; CVD, cardiovascular disease; DBP, diastolic blood pressure; DM, diabetes mellitus; HbA1c, glycated haemoglobin; IFCC, International Federation of Clinical Chemistry; SA, South Asian; SBP, systolic blood pressure

Patterns of glucose-lowering treatment

Overall, 36.2% of individuals remained on noninsulin monotherapy for the entire study period, 26.0% intensified from noninsulin monotherapy to noninsulin combination therapy, and 6.8% intensified from monotherapy to combination therapy to insulin therapy. Individuals with treatment patterns differing from guidelines included those who initiated with noninsulin combination therapy (3.5%) or insulin therapy (1.2%), those who intensified from noninsulin monotherapy directly to insulin (1.8%), and those who initiated with combination therapy and intensified to insulin (1.0%); 23.8% of the study population did not initiate any glucose-lowering treatment during the study period (Fig 2).

Fig 2

Diabetes therapy intensification sequence from diagnosis to end of follow-up by ethnic group. NIAD, noninsulin antidiabetic drug.

The ordering of glucose-lowering drug classes was comparable between ethnic groups at initiation but diverged in later stages. In all ethnic groups, the most popular first-line treatment was metformin (82%), the most popular second-line treatment was sulfonylureas (50%), and the most popular third-line treatments were newer-generation agents (34%). The choice for the fourth and fifth additional agent differed substantially by ethnic group; whereas South Asian and black groups were most likely to receive an additional newer-generation agent in both stages, white groups were more likely to receive insulin (Fig 3).

Fig 3

Drug classes by intensification stage and ethnic group.

DPP4i, dipeptidyl peptidase-4 inhibitor; GLP-1, glucagon-like peptide-1; SA, South Asian; TZD, thiazolidinedione.

Median time to initiation of noninsulin monotherapy was 3.2 months. Median time to intensification to noninsulin combination therapy was 28.5 months (2.4 years), and median time to intensification to insulin therapy was 45 months (3.8 years).

After adjusting for age, sex, deprivation, comorbidities, baseline HbA1c, BMI and smoking status, count of consultations and medications, calendar year, and clustering by practice, time to initiation was 21% faster in South Asian groups (95% CI 8%–36%, p < 0.001) and 29% faster in black groups relative to white (95% CI 5%–59%, p = 0.017). In contrast, time to intensification with noninsulin combination therapy was significantly slower for both nonwhite ethnic groups relative to white (South Asian HR 0.80, 95% CI 0.74–0.87, p < 0.001; black HR 0.79, 95% CI 0.70–0.90, p < 0.001). Ethnic differences widened further for intensification to insulin therapy, with South Asian groups taking twice as long to intensify as white groups (HR 0.49, 95% CI 0.41–0.58, p < 0.001) and black groups taking 31% longer to intensify (HR 0.69, 95% CI 0.53–0.89, p < 0.001) (Table 2, S3 Fig).

Table 2

Time to antidiabetic treatment initiation and intensification.

Initiation and intensification characteristics	Initiation of noninsulin monotherapy			Intensification to noninsulin combination therapy			Intensification to insulin therapy
N eligible to initiate/intensify		146,693			113,518			60,108
	White	South Asian	Black	White	South Asian	Black	White	South Asian	Black
N eligible to initiate/intensify	136,540	7,247	2,906	105,025	6,224	2,269	55,872	3,097	1,139
Percent who initiate/intensify at any time	98.9%	99.4%	97.8%	46.2%	42.8%	39.0%	20.9%	14.1%	16.2%
Duration of diabetes at start of follow-up period (years, mean, SD)	0	0	0	1.00 (1.99)	0.72 (1.63)	0.54 (1.47)	2.97 (3.05)	2.90 (3.05)	2.30 (2.92)
Time to treatment initiation/intensification
Months to initiation/intensification (mean, SD)	6.2 (33.5)	2.3 (17.4)	1.2 (15.4)	29.6 (45.1)	29.3 (43.6)	26.7 (42.6)	45.5 (59.8)	47.5 (64.5)	42.7 (60.6)
Relative risk versus white (HR, 95% CI, p-value)	1	1.21 (1.08–1.36) <0.001	1.29 (1.05–1.59) 0.017	1	0.80 (0.74–0.87) <0.001	0.79 (0.70–0.90) <0.001	1	0.49 (0.41–0.58) <0.001	0.69 (0.53–0.89) 0.004
Between-treatment characteristics
Mean HbA1c at diagnosis/start of follow-up period (%)	8 (2.1)	8.1 (2)	8.4 (2.4)	8.6 (1.9)	8.5 (1.9)	8.9 (2.3)	8.9 (1.8)	9 (1.9)	9.6 (2.3)
Mean HbA1c closest to date of initiation/intensification (%)	8.6 (2)	8.6 (1.9)	8.9 (2.3)	8.8 (1.7)	8.9 (1.8)	9.3 (2.1)	10 (1.9)	10.1 (1.9)	11 (2.5)
Number of HbA1c measurements between treatment stages (mean, SD)	1.5 (2.9)	1.1 (2.3)	0.8 (2.1)	1.7 (3.1)	1.2 (2.3)	.9 (2.1)	1.4 (2.5)	1 (1.9)	0.8 (2.1)
HbA1c count (RR, 95% CI, p-value)	1	0.94 (0.90–0.98) 0.002	0.90 (0.83–0.98) 0.017	1	0.78 (0.70–0.85) <0.001	0.72 (0.60–0.87) <0.001	1	0.74 (0.63–0.87) <0.001	0.64 (0.50–0.82) <0.001
Number of consultations between treatment stages (mean, SD)	12 (27.6)	9 (20.7)	6.9 (19)	14.8 (29.9)	10 (21.4)	8 (20.6)	11.3 (23.5)	7.9 (17.8)	6.4 (16.4)
Consultation (RR, 95% CI, p-value)	1	0.98 (0.94–1.02) 0.329	0.98 (0.93–1.04) 0.541	1	0.89 (0.78–1.01) 0.071	0.82 (0.66–1.01) 0.062	1	0.63 (0.52–0.76) <0.001	0.77 (0.59–1.01) 0.061

Population eligible for initiation excludes those on any diabetes drug in 90 days prior to diagnosis. All models adjusted for age, sex, deprivation, year of diagnosis, HbA1c, BMI, micro- and macro vascular comorbidities, depression, consultation count, smoking status and medication count at start of follow-up period, and clustering by practice. Models for intensification to combination therapy and insulin additionally account for time since diagnosis. Mean HbA1c and BMI taken as the latest in the 6 months prior to diagnosis (for model 1), initiation (for model 2), and intensification 1 (for model 3).

Abbreviations: BMI, body mass index; HbA1c, glycated haemoglobin; RR, rate ratio

In all treatment stages, nonwhite ethnic groups had fewer HbA1c measurements than white groups after adjusting for all confounders (Table 2).

From the total study population, 79,720 individuals with at least 12 months of follow-up following their first raised HbA1c value of >7.5% (58 mmol/mol) were included in the analysis of treatment inertia (S4 Fig).

Of all individuals who were treatment naive at diagnosis, 18% experienced therapeutic inertia when initiating treatment with noninsulin monotherapy. After accounting for all hypothesised confounders, no ethnic differences in therapeutic inertia were evident (South Asian odds ratio [OR] 0.97, 95% CI 0.76–1.24, p = 0.796; black OR 0.94, 95% CI 0.69–1.28, p = 0.683).

Of all individuals on noninsulin monotherapy, 68% experienced therapeutic inertia when intensifying to noninsulin combination therapy. Both South Asian and black groups experienced greater therapeutic inertia than white groups at this stage (South Asian OR 1.45, 95% CI 1.23–1.70, p < 0.001; black OR 1.43, 95% CI 1.09–1.87, p = 0.010).

Over 93% of all individuals on noninsulin combination therapy experienced therapeutic inertia when intensifying to insulin therapy. At this stage, ethnic minority groups were substantially more likely to experience therapeutic inertia than white groups (South Asian OR 2.68, 95% CI 1.89–3.80, p < 0.001; black OR 1.82, 95% CI 1.13–2.92, p = 0.013) (Table 3).

Table 3

Ethnic differences in therapeutic inertia (failure to intensify treatment within 12 months of HbA1c >7.5%).

Treatment stage	Ethic group	N with any HbA1c >7.5% and 12 months’ follow-up	Percent experiencing treatment inertia at 12 months, % (n)	Months between first HbA1c >7.5% and intensification/end f-up, mean (SD)	Odds of treatment inertia
Initiation of noninsulin monotherapy	White	34,546	18.4 (6,354)	6.9 (13.8)	1
	South Asian	1,683	16.2 (273)	6.3 (13)	0.97 (0.76–1.24), 0.796
	Black	560	17.5 (98)	6.1 (13.2)	0.94 (0.69–1.28), 0.683
Intensification to noninsulin combination therapy	White	54,307	67.1 (36,431)	29.2 (30)	1
	South Asian	3,076	69.4 (2,135)	29.8 (28.9)	1.45 (1.23–1.70), <0.001
	Black	1,029	68.7 (707)	31 (31.4)	1.43 (1.09–1.87), 0.010
Intensification to insulin therapy	White	36,480	93 (33,920)	54.8 (38.7)	1
	South Asian	2,061	96.2 (1982)	61.3 (43.1)	2.68 (1.89–3.80), <0.001
	Black	702	94.4 (663)	57.8 (42)	1.82 (1.13–2.92), 0.013

Regression model adjusts for age, gender, deprivation, HbA1c value, BMI value, smoking status, macro- and microvascular comorbidities, and depression at start of follow-up, number of consultations, and medications at the start of each follow-up period, calendar year at follow-up start, and clustering by practice. Models for intensification to combination therapy and insulin additionally account for time since diagnosis.

Abbreviations: BMI, body mass index; HbA1c, glycated haemoglobin

Secondary analyses

As in the primary analysis, no ethnic differences in the odds of experiencing therapeutic inertia were evident for initiation of noninsulin monotherapy when using 6.5% (48 mmol/mol) as the cutoff for raised HbA1c instead of >7.5% (58 mmol/mol) (South Asian OR 0.86, 95% CI 0.67–1.12, p = 0.274; black OR 0.86, 95% CI 0.60–1.23, p = 405) (S5 Table). Furthermore, the ethnic breakdown of individuals eligible for inclusion in the analysis of therapeutic inertia was comparable to the ethnic breakdown of individuals excluded due to insufficient follow-up after their initial raised HbA1c measure (S6 Table, full models for all analyses are in S7 Table).

Discussion

To our knowledge, this is the first UK-based study to examine the extent to which the appropriate and timely prescribing of diabetes treatment among individuals with type 2 diabetes differs by ethnic group. We report two main findings: Firstly, despite initiating noninsulin monotherapy more quickly after diagnosis than white groups, South Asian and black groups were more likely to experience delays when intensifying to noninsulin combination therapy and insulin therapy. Secondly, upon having uncontrolled HbA1c identified by the healthcare provider, South Asian and black groups were more likely to experience therapeutic inertia when intensifying to combination and insulin therapy.

Although several previous studies have quantified timeliness of initiation and intensification of antidiabetic medication in UK and international populations, ours is the first to identify differences by ethnic group. Similarly, although the literature describing ethnic differences in diabetic outcomes is extensive, we propose a mechanism via which these inequalities might occur—namely, via delayed intensification of treatment and therapeutic inertia following the identification of uncontrolled risk. Recognising this mechanism enables us to identify several priority target areas for reducing ethnic inequalities in the long-term management of diabetes in primary care settings.

Our data suggest that excessive delays in treatment intensification in ethnic minority populations may result from poorer monitoring of risk factors in these populations; in our study, both South Asian and black groups received fewer HbA1c measurements than white groups prior to intensification with both noninsulin combination therapy and insulin.

A recent systematic review of 53 studies worldwide has highlighted the widespread problem of therapeutic inertia in diabetes—with delays common across all stages of treatment initiation and intensification, though most pronounced around the time of intensification to insulin [22,26]. Correspondingly, we found that 67% of those intensifying to noninsulin combination therapy and 93% of those intensifying to insulin therapy experienced treatment inertia.

Strengths and weaknesses of this study

The strengths and limitations of routine EHRs for the purposes of diabetes research have been comprehensively outlined in a 2017 review [27]. This study benefitted from a large sample size drawn from a UK primary care database of over 15 million individuals, known to be representative of the UK population with respect to age, sex, and ethnicity [17]. This allowed for well-powered comparisons between the three main ethnic groups in the UK. The sample size of this study was significantly larger than those used in other recent UK and European studies; our study included over 160,000 individuals, compared with 2,500 to 24,000 reported elsewhere [23,24,28].

Furthermore, the crude time to intensification with insulin reported in our study (3.8 years) matched those of the UK-based cohort studies, which reported median times to insulin initiation between 3.2 and 4.9 years, indicating consistency of data quality and representativeness of the target population [24,28].

Thanks to the quality and outcomes framework, recent improvements in the completeness and quality of both diabetes and ethnicity data have facilitated robust examinations of ethnic differences in the management of type 2 diabetes in primary care settings [19,29]. Diagnoses of T2DM were ascertained using a validated algorithm designed to minimise miscoding and misclassification of diabetes type, reducing the likelihood that individuals with type 1 diabetes or other forms of diabetes were included in our study population [18]. Restriction of the study sample to individuals with at least 12 months of continuous registration prior to their initial diagnosis of T2DM ensured that diagnoses were truly incident and that a sufficient look-back period to capture key baseline covariates was present.

By restricting the analyses to white, South Asian, and black African/Caribbean groups, we were able to make clinically relevant comparisons between well-defined populations with distinct biological, sociocultural, and demographic characteristics, which can be meaningfully characterised as ethnicity [30]. Linkage to area-level deprivation data enabled us to separate the influences of ethnicity and deprivation, which are often conflated when examining health disparities.

General practice characteristics, such as size and participation in local enhanced service schemes, have been found to play a large role in observed variations in the quality of diabetes care [31]. By accounting for the clustering of individuals within general practices, we were able to appropriately account for the influence of practice-level factors on ethnic disparities.

Several limitations may have influenced our findings: Firstly, as EHRs are primarily used for patient care rather than research, data quality and completeness can vary significantly depending on the time period, disease area, and indicator of interest. These issues were mitigated by using data on a chronic disease condition managed predominantly in primary care settings [32,33] and by adjusting for calendar time in the analysis to account for secular trends in prescribing guidelines and treatment availability.

Secondly, the length of follow-up may have been insufficient to adequately characterise individuals intensifying with insulin therapy. Although 26.5% of the study population intensified from noninsulin monotherapy to noninsulin combination therapy during the follow-up period, only 5.8% of the population underwent all three intensification stages.

Furthermore, our definition of therapeutic inertia may have been too crude when looking at intensification to insulin therapy because individuals may have switched between various noninsulin combinations for some time before up-titrating with insulin therapy. As shown in our descriptive analysis, South Asian and black groups most commonly added noninsulin therapies as their fourth- and fifth-line treatments, whereas white groups most frequently added insulin.

Although the CPRD captures prescriptions made by the general practitioner, it does not hold any information on dispensing data; thus, we cannot know for certain whether prescriptions were filled or taken by the individual. Finally, we were unable to explore ethnic differences in rates of nonattendance to planned appointments, which may be related to ethnic differences in adherence to recommended therapeutic regimes and compliance with diabetes management plans.

Implications for clinicians and policymakers

The findings of this study highlight clear ethnic disparities in the long-term therapeutic management of type 2 diabetes, which likely contribute, at least in part, to the worse outcomes seen in ethnic minority populations in the UK. Given that baseline HbA1c prior to treatment initiation was higher in the ethnic minority than the white groups, our finding of faster initiation of noninsulin monotherapy in nonwhite groups is commensurate with the more advanced disease at diagnosis in the former groups. It may also reflect a preference for initial pharmacological (as opposed to lifestyle) interventions in ethnic minority groups, possibly indicating greater awareness of the disproportionate risk of major vascular outcomes in South Asian and black populations. Correspondingly, our previous study examining ethnic differences in management of type 2 diabetes around the time of diagnosis found that South Asian and black groups were offered nonpharmacological interventions (structured diabetes management and risk assessments) more quickly than white groups [34].

Reasons for treatment delays can stem from the healthcare system, the practitioner, and the patient. In the UK, these barriers include competing demands on practitioner time, financial constraints of the NHS (particularly in relation to the costs of newer medications), patient adherence, and concerns over side effects [35]. Intensification to insulin is particularly challenging because of the complexity of administration, the level of instruction required, and patient concerns around the use of injectable treatments [36]. Such challenges may be further exacerbated by cultural and language barriers on both the patient [37,38] and provider side [39], potentially explaining excessive delays in treatment intensification among ethnic minority groups. Ethnic differences in health beliefs and attitudes toward medication may also play a role in therapeutic inertia. A recent study from the United States found that statin undertreatment among African American groups was partially explained by lower levels of trust in healthcare practitioners and lower perceived safety of statins in the African American population compared with the white population [40]. Similarly, a UK survey of the Bangladeshi population found that refusal of insulin treatment was associated with fear of premature death, fear of weight gain, loss of independence, and lack of perceived improvements to quality of life [38].

Considering the wider spectrum of cardiometabolic disease, therapeutic inertia remains a substantial problem in the management of blood pressure and lipids. As with diabetes, causes for inertia in other disease areas are multifactorial, stemming from patient and provider, healthcare system, and policy factors. In addition to the important role of patient beliefs and understanding of health conditions and treatment options, other studies have highlighted the importance of education around clinical guidelines for both clinicians and patients, the use of multidisciplinary clinical teams, and the importance of quality improvement initiatives such as pay-for-performance schemes [41,42].

A final consideration is that treatment inertia may be appropriate for certain populations. As treatment decisions are increasingly made jointly between individuals and their care providers, purposeful therapeutic inertia may reflect shared concerns around frailty, treatment burden, and competing health-related priorities [43,44]. Risk of hypoglycaemia increases significantly with age and may be a key consideration in delaying treatment intensification in older age groups [45]. It is likely that some of the delays evident in our study population may be the result of individualised target setting to avoid overtreatment, particularly around the time of insulin intensification. However, even after accounting for burden of comorbidities, age, HbA1c, BMI, and medications, we found compelling evidence that ethnic minority groups nevertheless experience treatment inertia to a far greater degree than the white population.

Unanswered questions and future research

The first question arising from this work is to determine the relationship between prescribing patterns, long-term glycaemic control, and ethnic differences in micro- and macrovascular outcomes. Secondly, though difficult to quantify in EHRs, determining ethnic differences in adherence to diabetes treatment may shed light on how best to support specific population groups in maintaining good glycaemic control over the longer term. Thirdly, examining ethnic differences in the comparative effectiveness of different treatment regimes with respect to cardiovascular outcomes will be key to developing tailored treatment guidelines for multiethnic populations. Replicating and extending trials such as these to determine whether the risks and benefits of these treatments manifest differently between ethnic groups in real-world settings will form essential next steps toward the personalisation of care in populations with type 2 diabetes.

Transparency statement

Rohini Mathur is the manuscript’s guarantor. She affirms that the manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as originally planned have been explained.

Related manuscripts

The authors do not have related or duplicate manuscripts under consideration or accepted for publication elsewhere.

RECORD checklist.

RECORD, Reporting of Studies Conducted Using Observational Routinely Collected Data.

(DOCX)

Click here for additional data file.

Algorithm to assign ethnicity to study participants.

(TIFF)

Click here for additional data file.

Directed acyclic graph of the hypothesised relationship between ethnicity, therapeutic inertia, and intermediate confounders.

(TIFF)

Click here for additional data file.

Cumulative survival plots for survival analysis by ethnic group.

(TIFF)

Click here for additional data file.

Population flowchart for analysis of therapeutic inertia.

(TIFF)

Click here for additional data file.

ISAC scientific protocol.

ISAC, Independent Scientific Advisory Committee.

(DOCX)

Click here for additional data file.

Code lists for all study variables.

(DOCX)

Click here for additional data file.

Ethnic breakdown of study population according to the 16 categories of the UK census.

(DOCX)

Click here for additional data file.

Baseline characteristics for mixed/other and unknown groups compared with white.

(DOCX)

Click here for additional data file.

Time to diabetes treatment initiation and intensification for white versus other, mixed, and unknown ethnic groups.

(DOCX)

Click here for additional data file.

Therapeutic inertia at initiation of noninsulin monotherapy using a cutoff of 6.5% for definition of raised HbA1c.

HbA1c, glycated haemoglobin.

(DOCX)

Click here for additional data file.

Ethnic breakdown of individuals included and excluded from analysis of therapeutic inertia.

(DOCX)

Click here for additional data file.

Full models for time to initiation/intensification and therapeutic inertia.

(DOCX)

Click here for additional data file.

29 Dec 2019

Dear Dr. Mathur,

Thank you very much for submitting your manuscript "Ethnic differences in initiation and intensification of antidiabetic therapy and therapeutic inertia in adults with type 2 diabetes: A cohort study in the UK Clinical Practice Research Datalink" (PMEDICINE-D-19-04094) for consideration at PLOS Medicine.

Your paper was discussed among the editorial team and sent to independent reviewers, including a statistical reviewer. The reviews are appended at the bottom of this email and any accompanying reviewer attachments can be seen via the link below:

[LINK]

In light of these reviews, we will not be able to accept the manuscript for publication in the journal in its current form, but we would like to invite you to submit a revised version that fully addresses the reviewers' and editors' comments. You will appreciate that we cannot make a decision about publication until we have seen the revised manuscript and your response, and we expect to seek re-review by one or more of the reviewers.

In revising the manuscript for further consideration, your revisions should address the specific points made by each reviewer and the editors. Please also check the guidelines for revised papers at http://journals.plos.org/plosmedicine/s/revising-your-manuscript for any that apply to your paper. In your rebuttal letter you should indicate your response to the reviewers' and editors' comments, the changes you have made in the manuscript, and include either an excerpt of the revised text or the location (eg: page and line number) where each change can be found. Please submit a clean version of the paper as the main article file; a version with changes marked should be uploaded as a marked up manuscript.

In addition, we request that you upload any figures associated with your paper as individual TIF or EPS files with 300dpi resolution at resubmission; please read our figure guidelines for more information on our requirements: http://journals.plos.org/plosmedicine/s/figures. While revising your submission, please upload your figure files to the PACE digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at PLOSMedicine@plos.org.

We hope to receive your revised manuscript by Jan 17 2020 11:59PM. Please email us (plosmedicine@plos.org) if you have any questions or concerns.

***Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.***

We ask every co-author listed on the manuscript to fill in a contributing author statement, making sure to declare all competing interests. If any of the co-authors have not filled in the statement, we will remind them to do so when the paper is revised. If all statements are not completed in a timely fashion this could hold up the re-review process. If new competing interests are declared later in the revision process, this may also hold up the submission. Should there be a problem getting one of your co-authors to fill in a statement we will be in contact. YOU MUST NOT ADD OR REMOVE AUTHORS UNLESS YOU HAVE ALERTED THE EDITOR HANDLING THE MANUSCRIPT TO THE CHANGE AND THEY SPECIFICALLY HAVE AGREED TO IT. You can see our competing interests policy here: http://journals.plos.org/plosmedicine/s/competing-interests.

Please use the following link to submit the revised manuscript:

https://www.editorialmanager.com/pmedicine/

Your article can be found in the "Submissions Needing Revision" folder.

To enhance the reproducibility of your results, we recommend that you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see http://journals.plos.org/plosmedicine/s/submission-guidelines#loc-methods.

Please ensure that the paper adheres to the PLOS Data Availability Policy (see http://journals.plos.org/plosmedicine/s/data-availability), which requires that all data underlying the study's findings be provided in a repository or as Supporting Information. For data residing with a third party, authors are required to provide instructions with contact information for obtaining the data. PLOS journals do not allow statements supported by "data not shown" or "unpublished results." For such statements, authors must provide supporting data or cite public sources that include it.

Please let me know if you have any questions. Otherwise, we look forward to receiving your revised manuscript soon.

Sincerely,

Richard Turner, PhD

Senior Editor, PLOS Medicine

rturner@plos.org

-----------------------------------------------------------

Requests from the editors:

Please confirm that author REF was not an employee of Boehringer when the study was being carried out, and that the company had no involvement in planning or carrying out the study.

Please add the start and end years of the study to your title.

To the "methods and findings" subsection of your abstract, please add additional demographic information on study participants: i.e., mean age and proportion male.

Please add a few words to the abstract to summarize the approach to adjustment.

We ask you to add p values alongside 95% CI where available.

Please remove elements of repetition from the presentation of your findings, e.g., "... Treatment inertia was higher in south Asian and Black groups at this stage relative to White participants (south Asian OR 1.72 ...; Black OR 1.61 ...)".

Please add a new final sentence to the "methods and findings" subsection of your abstract to summarize the study's main limitations.

After the abstract, we ask you to add a new and accessible "author summary" section in non-identical prose. You may find it helpful to consult one or two recent PLOS Medicine research papers to get a sense of the preferred style.

Early in the methods section of your main text, please state whether the study had a protocol or prespecified analysis plan (and if so attach the document[s] as supplementary files, referred to in the text). Please highlight analyses that were not prespecified.

Please move the statement on ethics approval to the methods section of the main text.

Please avoid claims of "the first" (e.g., in the first paragraph of the discussion section), and where necessary add "to our knowledge" or similar.

In the discussion section of your main text, please restructure the text to create a discrete paragraph discussing limitations and add "This study had some limitations ..." or similar at the start.

Throughout your text, please reformat reference call-outs as follows: "... ethnic origin [1-3].".

Please substitute "sex" for "gender" as appropriate, throughout the paper.

Please check all reference citations to ensure that they match journal style. Several references appear to need additional access information (e.g., reference 33); reference 31 seems to lack a journal name; reference 8 needs reformatting. Please remove all iterations of "[Internet]".

Please revise the attached RECORD checklist so that individual items are referred to by section (e.g., "Methods") and paragraph number rather than by page or line numbers, as the latter generally change in the event of publication.

Please refer to the checklist in the methods section of your main text.

Comments from the reviewers:

*** Reviewer #1:

Thanks for the opportunity to review your manuscript. My role is as a statistical reviewer so my comments are focused on the study design, data, and analysis (and the presentation of these). I appreciate that detailed appendices on the derivation of the ethnicity information and the types of pharmacotherapy (and the other exposure information). This had headed off many of my usual questions I have from studies using GPRD and other similar routinely collected data sources, and will prove to be useful to anyone else attempting to do something similar to your own study. This study uses UK routinely collected data to estimate whether initiation and intensification of pharmacotherapy for T2DM differs according to ethnicity. From my own experience doing similar work I appreciate how complicated an analysis like this is, and I think that you have presented the details of this very clearly. I have some general queries around the study, and some more specific queries and points following that.

General

Was a study protocol or statistical analysis plan developed for this study? It would helpful to see this if it was develop a-priori.

Will diagnoses and pharmacotherapy provided by specialists (endocrinologists) appear in the dataset, and medicine given during hospital stays?

One of the key parts of the data that this study hinges on is accurate identification of patient's ethnicity, and the algorithm used to do this is clearly presented. This seems reasonable, and rather than give you 'death by sensitivity analyses' asking for the main results to be repeated by the various assumptions that could be made here, is it possible to quantify what the frequency of ethnicity in the study would change if a stricter definition was used where all records had to be consistent? Has there been any methods papers on the UK CPRD looking at the quality of the use of this variable? My own experience with other sources of routinely collected data is that this can wildly vary - although it certainly can be well recorded and reliable.

There is a fair amount of literature looking at T2DM management in South-East Asians in other countries - in this study these were put together into the 'other' category of ethnicity. Was there an insufficient number of self-identifying south-east Asian persons to consider these in this study? Is it possible to have a table that reports on the frequencies of the persons excluded as 'other' according to the 16 category ethnicity variable in the appendix?

Has the use of self-identification to an ethnic group changed over time in the CPRD? Could you clarify how someone would be classified if they had never had valid record of ethnicity in the data?

Specific queries

P6, Paragraph 2. What are the adjustments to the BMI categories?

P6, Paragraph 3. How are participants who move from one form of therapy to a less intense form of therapy, and then to another form of therapy treated? i.e. Monotherapy to none, none to combination therapy? Is the CPRD based on prescription or dispensing records?

P7. Paragraph 4. Are non-pharmacotherapy periods considered to be a 'drug regime'?

P7, Paragraph 5. What were the levels in the multilevel Cox model? Is this to account for multiple treatment periods (i.e. none -> mono, mono -> combination) within the same individual? [I've noticed this is clarified below, but it would be helpful to explain the levels here as well] Were proportional hazards between levels of covariates checked, and how was this done?

P8, Paragraph 3. If a covariate (e.g. microvascular complications) changes during a treatment period, how is this accounted for?

P8, Paragraph 4. Were there any checks of overdispersion with the Poisson regression model?

P11, Paragraph 2/3. I'd also like to see cumulative incidence plots of this if possible (in an appendix is fine), and the median time to intensification (i.e. the time when 50% of those able to receive intensification received it.

*** Reviewer #2:

The manuscript by Mathur et al deals with a very relevant issue regarding ethnic differences in the implementation of antidiabetic treatment in real-world practice. The authors used the well-known CPRD database to explore the potential ethnic differences in initiation and intensification of hypoglycemic therapy in type 2 diabetes in the UK. The methodology used is sound and the paper well written and easy to read. The findings are relevant for daily clinical practice and for future research in this field.

This reviewer would like the authors to address the following issues:

- In the Abstract, under Conclusions, the authors include a statement on delays in treatment intensification (second sentence). Actually, delays may be related to many factors and I would recommend rephrasing the sentence or delete it.

- Under Methods, it is not clear whether the information available on drug use is based on prescription by physicians or on pharmacy dispensation. This is relevant as information on the latter is a better proxy to adherence. In addition, this is an issue that deserves a comment in the discussion.

- Also, under Methods, please explain the method used to handle missing data.

- In Statistical analysis, section on 'Time to treatment initiation and intensification',

- In the section on Results, first paragraph, the sum of percentages provided sum up to 100.1%. Please , check this. Also, please, check the percentages given under 'Patterns of glucose-lowering treatment'.

- Although the duration of diabetes is most probably not different among groups, this information is of interest for the reader.

- The authors included the number of consultations which is different between groups. However, it is important to know whether the difference may be due to non-attendance to the planned appointments. Do the authors have information on adherence to planned follow-up consultations. This may be also an indicator of adherence to the therapeutical plan; this may clearly influence the outcomes of the study.

- This reviewer understands that the data analyzed are those that correspond to primary care diabetes management. Do the authors have information on access of the subjects to secondary/tertiary care of the different ethnic groups? Were any subjects managed at this level for their diabetes?

- There are several factors known to affect the outcomes measured in this study, both on the patient's and on the professional's side. One of them is on the subject with diabetes, i.e. depressive disorders; do the authors have data on differences among ethnic groups in the prevalence of this co-morbidity? This would be relevant as a confounding variable.

- Other confounding variables known to affect the outcomes are cultural differences in perception of the disease, perception of the impact on daily life (especially for insulin) and communication barriers between patients and health care professionals (very likely to affect treatment implementation). These variables have been shown to impact on treatment outcome and probably deserve a comment in the discussion.

- In the Discussion, under 'Unanswered questions and future research', I do not see the reason to specifically mention the CAROLINA trial. Please, explain.

*** Reviewer #3:

Thank you for the opportunity to review this review this article which uses the UK primary care electronic healthcare records to assess the association between Type 2 diabetes glucose lowering treatment intensification and ethnicity. The article shows that despite similar initial treatment intensification there appears to be substantially more inertia in initiating later glucose lowering therapies in those of Black and South Asian ethnic minority background, associated with reduced follow up and HBA1c monitoring.

The article is well written and addresses a question of some importance with (to my non methodologist eyes) robust analysis. However one factor notably missing is an attempt to understand/explain these differences, or relate them to findings in other disease areas.

The finding of differences later in care is at odds with that of initial treatment and it would be helpful to consider this further. Within the limitations of the dataset is there scope here to start to shed some light on cause of this variation, that might therefore help target action to address these imbalances? For instance how does this variation alter once reduced follow up/monitoring is taken into account - if a large effect this might point to differences in offering or attending following up being key areas to address, rather than patient or physician willingness to accept or prescribe more intense therapy. Is there scope to assess adherence (at least in terms of prescription pick up)? Poor adherence to existing therapy can be a reason for not adding additional therapy. It may also be informative to break down the effect adjusting for confounders in the same way as other authors in related fields have done (for example see figure 3 of Nanna et al JAMA cardiology 2018 PMID: 29898219).

On brief search it appears to me that there may be substantial related work in other fields relevant to this exact issue including in management of hypertension and treatment of lipids - it is unlikely that an issue with titrating diabetes therapy exists (or is best addressed) in isolation, so it would be helpful to at least put this in context of some of the work in other conditions in the discussion. It would also be helpful comment on possible explanation and related work - for example differences in statin use by ethnicity appear to be heavily influenced by health beliefs (see Nanna et al article above) and a brief search suggests there has been previous qualitative work on this area in type 2 diabetes which may be relevant - for example there appears to be work around ethnicity and health beliefs relating to insulin initiation that is directly relevant to the findings here (e.g. PMID: 27574375 and references within). The substantial increased use of the new more expensive glucose lowering agents in later stages in those of non white ethnicity may be relevant to this.

Additional points

1. It is a little confusing exactly what has been adjusted for, and why. Baseline covariates (page 6) suggests factors like smoking and BMI were included as covariates in analysis, but many of these are absent from the related statement of a-priori confounders on page 8. Page 8 suggests to my mind that consultations and number of measurements were also adjusted for, but this does not appear to be the case in table 2. In table 3 all the things on page 6 again crop up. This leaves me rather lost, and also confused why different covariates may be considered (if indeed this is the case) for very closely related analysis of time to intensification of therapy and treatment inertia, which would appear to be aspect of the same thing. Related to this the DAG suggests the authors consider there is a similar relationships for hbA1c and BMI/smoking etc but (if page 8/table 2 is correct) these appear to be treated differently in analysis. One would expect BMI in particularly to affect choice of agent (given clinicians and patients may be more reluctant to use medications associated with weight gain such as insulin/sulfonylureas where there is obesity).

2. There is to my mind a possibility of residual confounding here that may be exacerbated by the use of adjustment for artificial subgroups (as the covariates description on page 6 suggests) rather than using continuous covariates. Was there a reason to split age into deciles rather than adjust for age as a continuous covariate (in the same way as other groups using this dataset have done?), ditto for BMI which is even more crudely categorised - a BMI of 30 and 60 are likely to have different influences on choice of treatment.

3. We know there is marked regional variation in use of newer agents and potentially rural/urban differences in care. There will also be major differences in ethnicity by rural/urban status and region. I assume this will be accounted for by modelling at the practice level, could the authors reassure me that this is the case?

***

Any attachments provided with reviews can be seen via the following link:

[LINK]

7 Feb 2020

Submitted filename: PLOS_Med_Response_to_reviewers_05022020.docx

Click here for additional data file.

5 Mar 2020

Dear Dr. Mathur,

Thank you very much for re-submitting your manuscript "Ethnic differences in initiation and intensification of antidiabetic therapy and therapeutic inertia in adults with type 2 diabetes: A cohort study in the UK Clinical Practice Research Datalink" (PMEDICINE-D-19-04094R1) for consideration at PLOS Medicine.

I have discussed the paper with editorial colleagues, and it was also seen again by three reviewers. I am pleased to tell you that, provided the remaining editorial and production issues are dealt with, we expect to be able to accept the paper for publication in the journal.

The remaining issues that need to be addressed are listed at the end of this email. Any accompanying reviewer attachments can be seen via the link below. Please take these into account before resubmitting your manuscript:

[LINK]

Our publications team (plosmedicine@plos.org) will be in touch shortly about the production requirements for your paper, and the link and deadline for resubmission. DO NOT RESUBMIT BEFORE YOU'VE RECEIVED THE PRODUCTION REQUIREMENTS.

***Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.***

In revising the manuscript for further consideration here, please ensure you address the specific points made by each reviewer and the editors. In your rebuttal letter you should indicate your response to the reviewers' and editors' comments and the changes you have made in the manuscript. Please submit a clean version of the paper as the main article file. A version with changes marked must also be uploaded as a marked up manuscript file.

Please also check the guidelines for revised papers at http://journals.plos.org/plosmedicine/s/revising-your-manuscript for any that apply to your paper. If you haven't already, we ask that you provide a short, non-technical Author Summary of your research to make findings accessible to a wide audience that includes both scientists and non-scientists. The Author Summary should immediately follow the Abstract in your revised manuscript. This text is subject to editorial change and should be distinct from the scientific abstract.

We hope to receive your revised manuscript within 1 week. Please email us (plosmedicine@plos.org) if you have any questions or concerns.

We ask every co-author listed on the manuscript to fill in a contributing author statement. If any of the co-authors have not filled in the statement, we will remind them to do so when the paper is revised. If all statements are not completed in a timely fashion this could hold up the re-review process. Should there be a problem getting one of your co-authors to fill in a statement we will be in contact. YOU MUST NOT ADD OR REMOVE AUTHORS UNLESS YOU HAVE ALERTED THE EDITOR HANDLING THE MANUSCRIPT TO THE CHANGE AND THEY SPECIFICALLY HAVE AGREED TO IT.

Please ensure that the paper adheres to the PLOS Data Availability Policy (see http://journals.plos.org/plosmedicine/s/data-availability), which requires that all data underlying the study's findings be provided in a repository or as Supporting Information. For data residing with a third party, authors are required to provide instructions with contact information for obtaining the data. PLOS journals do not allow statements supported by "data not shown" or "unpublished results." For such statements, authors must provide supporting data or cite public sources that include it.

Please let me know if you have any questions. Otherwise, we look forward to receiving the revised manuscript shortly.

Sincerely,

Richard Turner, PhD

Senior Editor, PLOS Medicine

rturner@plos.org

------------------------------------------------------------

Requests from Editors:

We suggest a more compact title: "Ethnic disparities in initiation and intensification of diabetes treatment in adults with type 2 diabetes in the UK, 2004-2017: a cohort study"

Please add a full point as needed after "hypothesized confounders" in the abstract.

Please adapt "... groups were faster to initiate ..." (abstract) to "... non-insulin monotherapy was initiated earlier in south Asian and Black groups ..." or similar.

In the "Conclusions" subsection of your abstract, please remove the word "strong", bearing in mind the research design, or substitute "persuasive" or similar.

At the end of the "Introduction" section of your main text, are (b) and (c) different? Please reword as appropriate.

Please refer to the attached "RECORD" checklist at the appropriate point in your methods section.

Please remove the short section on "patient and public involvement".

Under "ethnic differences" (results, third paragraph), please correct "p<0.001".

In the first paragraph of the discussion, please adapt "... experienced significant delays" to "were more likely to experience delays" or similar.

Please avoid the phrase "white majority".

Please use the term "diabetes treatment", or similar, rather than "antidiabetic therapy" throughout the ms.

Please substitute "sex" for "gender" throughout the article.

Please reposition all reference call-outs as follows: "... adverse diabetes outcomes [12,13]. In the UK ...".

In the reference list, please format author names correctly where needed, e.g., references 32, 28 and 40.

Please add journal names where they are missing, e.g., references 18 and 44.

In table 1, please remove the colour highlighting of "depression".

Comments from Reviewers:

*** Reviewer #1:

Thank you for the revision and replies to the comments made on the original submission. Overall I consider my original comments to be resolved with the changes to the manuscript in this version and the inclusion of the protocol in the supplementary materials. The change to use continuous covariates (i.e. BMI) as suggested by reviewer 3 is a good change to the manuscript and the presentations of hazard/odds ratios in Table 2/3 highlights the differences between groups well. I had one query relating to the identification of ethnicity prompted by the additional table, and a comment about the figures in the main part of the paper.

Table S1 is very helpful to understand the selection of ethnic groups as the main exposure in the study. The majority of persons haven an 'unknown' ethnicity classification and are excluded from the analysis, which would presumably mean that many of the persons who are actually 'White' are excluded at this this step. Are the 'White' group very similar to those in the 'Unknown' category? i.e. are the differences seen between ethnicities in treatment intensification driven by selection bias on identification as 'White' in the CPRD? My own experiences with routinely-collected data on self-reported CoB suggest that people and health services that elicit this information from patients are likely to be different to those that do not. A table comparing the baseline characteristics of the excluded 'Other' with 'White would resolve whether a select group more likely to receive treatment intensification was used by only selecting the people with a confirmed 'White' ethnicity code.

Prior to publication, the Figures 2 and 3 should be revised to a higher standard: 1) there is not main axis title, 2) The numbers in Figure 2 are difficult to read against the striped background, and in the rarer intensification patterns they overlap and can't be read, 3) The table in Figure 3 needs some revision so it's clear there are 5 drug reported across the axis (darker gridlines between drug numbers?).

*** Reviewer #2:

The authors have properly addressed the issues raised by this reviewer

*** Reviewer #3:

Thank you for fully addressing my comments. I have only one further comment (for information): while there are clear limitations in assessing adherence in CPRD it is possible to assess whether a patient is requesting sufficient repeat prescriptions to cover the prescribed dose, and this measure has been shown to be associated with glycaemic control in CPRD - see Farmer et al Diabetes Care 2016 PMID: 26681714.

***

Any attachments provided with reviews can be seen via the following link:

[LINK]

14 Apr 2020

Submitted filename: Response_to_reviewers_6032020.docx

Click here for additional data file.

15 Apr 2020

Dear Dr. Mathur,

On behalf of my colleagues and the academic editor, Dr. Didac Mauricio, I am delighted to inform you that your manuscript entitled "Ethnic disparities in initiation and intensification of diabetes treatment in adults with type 2 diabetes in the UK, 1990-2017: a cohort study" (PMEDICINE-D-19-04094R2) has been accepted for publication in PLOS Medicine.

PRODUCTION PROCESS

Before publication you will see the copyedited word document (in around 1-2 weeks from now) and a PDF galley proof shortly after that. The copyeditor will be in touch shortly before sending you the copyedited Word document. We will make some revisions at the copyediting stage to conform to our general style, and for clarification. When you receive this version you should check and revise it very carefully, including figures, tables, references, and supporting information, because corrections at the next stage (proofs) will be strictly limited to (1) errors in author names or affiliations, (2) errors of scientific fact that would cause misunderstandings to readers, and (3) printer's (introduced) errors.

If you are likely to be away when either this document or the proof is sent, please ensure we have contact information of a second person, as we will need you to respond quickly at each point.

PRESS

A selection of our articles each week are press released by the journal. You will be contacted nearer the time if we are press releasing your article in order to approve the content and check the contact information for journalists is correct. If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximize its impact.

PROFILE INFORMATION

Now that your manuscript has been accepted, please log into EM and update your profile. Go to https://www.editorialmanager.com/pmedicine, log in, and click on the "Update My Information" link at the top of the page. Please update your user information to ensure an efficient production and billing process.

Thank you again for submitting the manuscript to PLOS Medicine. We look forward to publishing it.

Best wishes,

Richard Turner, PhD

Senior Editor

PLOS Medicine

plosmedicine.org