Screening for type 2 diabetes: do screen-detected cases fare better?

Abstract

Aims

To investigate whether diabetes cases detected through screening have better health outcomes than clinically detected cases in a population-based cohort of adults who were eligible to be screened for diabetes at 10 year intervals.

Methods

The Västerbotten Intervention Programme is a community and individual-based public health programme in Västerbotten County, Sweden. Residents are invited to clinical examinations which include screening for diabetes by oral glucose tolerance tests (OGTT) at age 30, 40, 50 and 60 years (individuals eligible for screening, n=142,037). Between 1992 and 2013 we identified n=1,024 screen-detected cases and n=8,642 clinically detected cases using registry data. Clinically detected cases were either prior screening participants (n=4,506) or people who did not participate in screening (non-participants, n=4,136). Cases were followed from date of detection until end of follow-up, emigration, death, or incident cardiovascular (CVD), renal disease or retinopathy event, and compared using Cox proportional hazard regression adjusted for calendar time, age at and year of detection, sex and socioeconomic status.

Results

Average age at diagnosis was 4.6 years lower for screen-detected cases compared to clinically detected cases. Overall, clinically detected cases had worse health outcomes than screen-detected cases, hazard ratio (HR) for all-cause mortality 2.07 (95% confidence interval (CI) 1.63, 2.62). Compared to screen-detected cases, all-cause mortality was higher for clinically detected cases who were screening non-participants (HR 2.31, 95% CI 1.82, 2.94) than for clinically detected cases who were prior screening participants (HR 1.70, 95% CI 1.32, 2.18). Estimates followed a similar pattern for CVD, renal disease and retinopathy.

Conclusions

Screen-detected diabetes cases were diagnosed earlier and appeared to fare better than clinically detected cases with regards to all-cause mortality, CVD, renal disease and retinopathy. How much of the association is explained by earlier treatment rather than healthy user, lead and length time biases warrants further investigation.

Introduction

More than an estimated one million adults in the UK and 160,000 adults in Sweden are living with undiagnosed diabetes [1], which is potentially detectable by screening. In screening for type 2 diabetes, one cluster-randomised controlled trial in a high risk UK population (ADDITION-Cambridge [2]) and a controlled trial in a high risk Danish population (ADDITION-Denmark [3]) found no effect on mortality in the population after approximately 10 years. One cohort study in an average risk UK population (the Ely cohort) reported a reduction of mortality in 1990-1999, but no effect 10 years later [4]. In the Ely cohort the average lead time for a diabetes diagnosis due to screening was estimated at 3.3 years, but this was not associated with lower incidence of adverse health outcomes for cases detected earlier through screening [5]. A study in Sweden compared diabetes cases detected in an opportunistic screening programme with cases detected clinically in the same eligible population and found no difference in age at diagnosis or any effect on health outcomes for screen-detected cases [6]. However, in ADDITION-Denmark, a lead time of 2.2 years was associated with lower mortality and cardiovascular disease (CVD) risk among cases in the screened group [7].

A review found that the positive predictive values of a single biochemical screening test for diabetes ranged between 24 and 48% [8], meaning that more than half of those with positive screening tests likely have only transient non-diabetic hyperglycaemia. Although it is known that compared to normoglycaemia, those with non-diabetic hyperglycaemia have an increased risk of CVD and death [9], the fate of those with unconfirmed diabetic test results specifically has not been studied.

Following reports in the 1970’s of relatively high CVD mortality in the Swedish county of Västerbotten, a community public health intervention programme was launched [10]. The Västerbotten Intervention Programme (VIP) was first implemented in 1985 and reached full coverage in 1992. There is some evidence that the overall public health programme has had a positive impact on all-cause and CVD mortality [11]. VIP has both a community, and individual focus with invitations to standardised health examinations in primary healthcare [10]. Crucially, these include oral glucose tolerance tests (OGTT), which allows us to study VIP as a model for an organised systematic universal diabetes population screening programme.

We aimed to investigate the association between screen-detection of type 2 diabetes and all-cause mortality, CVD events, renal disease, and retinopathy in this population-based cohort of adults eligible to be screened at 10 year intervals. The secondary aim was to investigate the rate of these outcomes in unconfirmed screen-positive cases.

Methods

The Västerbotten Intervention Program

Since 1985, residents of Västerbotten County are eligible to be invited to standardised health examinations at the age 30 (until 1995), 40, 50 and 60 years, as described in detail previously [10]. Briefly, at every VIP examination participants are asked to complete a comprehensive questionnaire that covers among other things lifestyle behaviour and health status, which together with OGTT results is used as basis for a motivational health promotion dialogue. Individuals found to have non-diabetic hyperglycaemia receive referrals for a follow-up visit with a nurse, and individuals found to have diabetes are referred to primary care for confirmatory testing and treatment [10]. Overall rate of participation at the first eligible VIP examination over the study period has ranged from 48 to 67% [12]. Objectively measured data, such as BMI and blood glucose measurements, and associated questionnaire data collected in the VIP health examinations may be linked to local and national registers using the Swedish personal identification number [13]. Ethical approval for this study was granted by the Regional Ethical Review Board, Umeå, Sweden [Dnr 2013-395-31M, Dnr 2014-410-32M].

Eligible study population

Eligible individuals (n=142,037) were identified in the Population Register which is maintained by the Swedish Tax Agency [14]; they were resident in Västerbotten County between 1992 and 2013, born between 1932 and 1971, aged 30 years or older, and had sufficient information available to enable record linkage to other population-based registers (Figure 1). We excluded individuals who had a record of prevalent diabetes (n=1,761, see details below), leaving a study population of 140,276 individuals eligible for screening among whom incident diabetes cases were identified. We followed up the study individuals for detection of diabetes from 1 January 1992, their 30^th birthday, or date of move to Västerbotten County, until 31 December 2013, death or emigration, whichever came first.

Figure 1.

Flowchart describing the study population. The prevalent diabetes-free study population (n=140,276) was eligible for screening; n=46,209 individuals did not participate in screening and n=94,067 participated. Among participants, n=16,214 had non-diabetic hyperglycaemia (n=1,898 later detected clinically) and n=3,483 screened positive for diabetes. Of those who screened positive, the diagnosis was confirmed within one year in 1,024 cases, and remained unconfirmed in 1,403 (n=280 cases were found to have diabetes other than type 2). N=776 cases were clinically detected after having screened positive for diabetes more than a year previously, n=1,716 cases were clinically detected after having normoglycaemia in screening and 116 cases were clinically detected after participating in a health examination without completing an OGTT; in total, n=4,506 cases were clinically detected among all screening participants. Among screening non-participants, 4,136 cases were clinically detected.

OGTT: Oral glucose tolerance test, VIP: Västerbotten Intervention Programme.

Screen-detected diabetes

In the study population, n=94,067 individuals attended at least one VIP examination, which corresponds to a participation rate of 67.1% (Figure 1). Each health examination included an OGTT using a 75g oral glucose load [15]. In total, n=16,214 individuals had a non-diabetic hyperglycaemic OGTT measurement (fasting or 2 hour capillary plasma blood glucose of 6.1-6.9 mmol/L or 8.9-12.1 mmol/L, respectively) and n=3,483 individuals had a diabetic OGTT measurement (defined according to current diagnostic criteria as a fasting or 2 hour plasma blood glucose level of ≥7.0 mmol/L or ≥12.2 mmol/L, respectively) without any prior record of a diabetes diagnosis. Among participants with diabetic OGTT measurements, n=1,024 (29.4%) cases were confirmed as type 2 diabetes cases within one year in at least one medical or prescription record (see details below); the majority were confirmed in either the Västerbotten County Medical Record system (n=486, 47.5%) or DiabNorth (n=425, 41.5%); n=1,403 (40.3%) individuals did not have any record of a diabetes diagnosis besides one diabetic OGTT and were thus unconfirmed screen-positive cases (Figure 1). In addition, n=776 (22.3%) individuals had a diabetic OGTT but were only confirmed as diabetes cases after more than 1 year and thus considered clinically detected diabetes cases (see below), and n=280 individuals had a medical record of another type of diabetes than type 2, e.g. type 1 diabetes, gestational diabetes (Figure 1).

Clinically detected diabetes

In total, n=8,642 diabetes cases were identified in five sources of medical and prescription records (numbers and percentages refer to cases with earliest date of diagnosis in the source): 1) DiabNorth (n=1,205, 13.9%), a register of validated diabetes diagnoses in VIP until 2012 [16]; 2) The National Diabetes Register (NDR) (n=603, 7.0%), a resource linked to primary care that was initiated in 1996 [17] (coverage of the NDR in Västerbotten County ranged between 50% to over 70% of diabetes cases registered [18]); 3) The Västerbotten County Medical Record system (n=4,984, 57.7%) which includes records of all primary care visits in the county since 2006 with partial coverage since 1993; 4) The Prescribed Drugs Register (n=896, 10.4%) which includes records of all dispensed prescription drugs since 1 July 2005 (diabetes record defined as dispensation of any drug with an Anatomical Therapeutic Chemical code A10) [19]; 5) The National Patient Register which includes all inpatient discharge records since 1987 (n=858, 9.9%) with partial national coverage since 1964, and outpatient records since 2001 (n=96, 1.1%) (diabetes defined as a record with a primary or contributory diagnosis of diabetes (International Classification of Diseases (ICD) codes 250 (revision 9) or E11, E13, E14 (revision 10)) [20]. These sources were also used to confirm screen-detected cases, see above. Using capture-recapture [21] we estimate that 96.9% of diabetes cases in the population were identified using a combination of these sources.

Clinically detected cases were divided into two groups based on participation in VIP screening before first detection of diabetes, VIP participants were further divided into three groups. N=4,506 (52.1%) diabetes cases had participated in VIP at least once before the date of detection of diabetes and had had either a 1) diabetic (if > 1 year pre-detection, n=776), 2) non-diabetic hyperglycaemic (n=1,898), or 3) normoglycaemic OGTT (n=1,716) (n=116 cases had participated in VIP but had missing OGTT data). The remaining n=4,136 (n=47.9%) clinically detected diabetes cases never participated in VIP screening, despite being eligible to do so, before detection of diabetes (Figure 1).

Events and outcomes

The primary outcome was date of death as identified by record linkage to The Total Population Register (Statistics Sweden) [14]. Secondary outcomes were dates of incident CVD events (myocardial infarction, heart failure, stroke or peripheral arterial disease), incident renal disease, or incident retinopathy. All disease events were identified in the National Patient Register as well as the Cause of Death Register which has complete coverage since 1961 [22]. See electronic supplementary material (ESM) Table 1 for a detailed list of all ICD-9 and ICD-10 codes used to classify events.

Table 1.

Descriptive statistics of type 2 diabetes mellitus cases, Västerbotten Intervention Programme 1992-2013.

							Incident clinically detected cases
	Total		Confirmed screen-detected cases		Unconfirmed screen-positive cases		All		Screening participants		Screening non-participants
	n	%	n	%	n	%	n	%	n	%	n	%
Totala	11,069	100.0	1,024	9.3	1,403	12.7	8,642	78.1	4,506	40.7	4,136	37.4
Men	6,459	58.4	595	58.1	783	55.8	5,081	58.8	2,572	57.1	2,509	60.7
Women	4,610	41.6	429	41.9	620	44.2	3,561	41.2	1,934	42.9	1,627	39.3
Age at detection
Mean (median, SD)	58.5	(59.9, 9.7)	55.1	(59.8, 6.4)	53.4	(59.7, 8.0)	59.7	(60.2, 9.9)	61.5	(62.2, 9.0)	57.9	(58.4, 10.4)
30-39	246	2.2	0	0.0	27	1.9	219	2.5	20	0.4	199	4.8
40-49	1,410	12.7	82	8.0	220	15.7	1,108	12.8	441	9.8	667	16.1
50-59	3,296	29.8	335	32.7	406	28.9	2,555	29.6	1,222	27.1	1,333	32.2
60-69	4,494	40.6	607	59.3	750	53.5	3,137	36.3	1,851	41.1	1,286	31.1
70-	1,623	14.7	-		-		1,623	18.8	972	21.6	651	15.7
Year of detection
1992-1999	2,012	18.2	217	21.2	369	26.3	1,426	16.5	352	7.8	1,074	26.0
2000-2006	3,841	34.7	369	36.0	537	38.3	2,935	34.0	1,426	31.6	1,509	36.5
2007-2013	5,216	47.1	438	42.8	497	35.4	4,281	49.5	2,728	60.5	1,553	37.5
Socioeconomic status in the 1990 censusb
Manual workers	5,516	49.8	479	46.8	701	50.0	4,337	50.2	2,259	50.1	2,089	50.5
Non-manual workers	4,235	38.3	411	40.1	543	38.7	3,285	38.0	1,737	38.5	1,534	37.1
Self-employed	744	6.7	82	8.0	103	7.3	562	6.5	305	6.8	264	6.4
Undefined	574	5.2	52	5.1	56	4.0	458	5.3	205	4.5	249	6.0
Prior CVD eventc	1,420	12.8	55	5.4	53	3.8	1,312	15.2	673	14.9	639	15.4

Unconfirmed screen-positive cases had only a diabetic screening result, whereas confirmed screen-detected cases had a diabetic screening result and a medical or prescription record of diabetes within one year. Clinically detected diabetes cases were identified in five sources of medical and prescription records, unrelated to screening.

^aPercentages across the row.

^bSocioeconomic status imputed for n=1,561 individuals who had missing information.

^cPrevious CVD events include those that coincide with date of diabetes detection (n=227).

CVD: Cardiovascular disease, SD: Standard deviation.

Other variables

Sex and date of birth were taken from the Total Population Register [14]. Socioeconomic status (SES) was categorised into four levels (manual workers, non-manual workers, self-employed and undefined) in the 1990 census. For n=1,561 cases who had missing information on SES we imputed values using single imputation (ice command in Stata) based on the variables sex, birthdate, VIP participation (yes/no) and international migration status. Prior CVD events were ascertained in the National Patient Register and defined as above if they occurred before or on the same date as diabetes detection. For VIP participants, height, weight, blood pressure and total serum cholesterol were objectively measured at VIP examinations [10]. Body mass index (BMI) was calculated as weight in kilograms divided by height in metres squared. The VIP questionnaire includes the question “How has your health been in the past year?”; we dichotomosed the response alternatives good, pretty good, somewhat good, pretty bad, and bad into overall good/overall bad [23].

Statistical analysis

Diabetes cases were followed up from date of first detection of diabetes (i.e. date of VIP screening for screen-detected cases or date of first diabetes record for clinically detected cases) until 31 December 2013, date of death, emigration, or date of outcome event depending on the model. Crude incidence (IR) and mortality rates (MR) were calculated as events divided by time at risk scaled to 1,000 person-years (PYR). Directly standardised IR (StdIR) and MR (StdMR) were calculated with weights derived from the follow-up time and age and sex distribution in the total study population (n=140,276) as reference. Associations between mode of detection of diabetes and outcomes were assessed using Cox proportional hazard regression generating hazard ratios and 95% confidence intervals. All models used calendar time as the underlying time-scale and were adjusted for age at and calendar year of diabetes detection as continuous variables, sex and SES in 1990. Differences in biomarkers measured at concurrent or previous VIP examination between screen-detected vs. unconfirmed screen-positive and clinically detected diabetes cases who were screening participants were tested using the t-test for continuous variables, and the χ²-test for categorical variables. To disentangle the contributions of early treatment and length time bias to the difference in outcome rates between screen-detected cases and clinically detected cases who had been screening participants, we conducted two sensitivity analyses: 1) main analysis further adjusted for prior CVD event status, and 2) main analysis further adjusted for prior CVD event status, and the following variables measured at concurrent or previous VIP examination: BMI, diastolic and systolic blood pressure, total serum cholesterol, self-reported health status, and time from screening to diabetes detection in years.

Results

We identified n=9,666 diabetes cases in total, constituting a cumulative incidence of 6.9% in the study population. Screen-detected diabetes cases were on average 4.6 years younger at diagnosis than clinically detected cases, 6.4 years younger than clinically detected cases who were screening participants, and 2.8 years younger than clinically detected cases who were screening non-participants (Table 1). There was a substantial difference in the proportion of individuals who had experienced a CVD event prior to their date of detection of diabetes between screen and clinically detected cases, 5.4% for screen-detected cases vs. 15.2% for clinically detected cases. Among all clinically detected diabetes cases, n=227 (2.6%) had a CVD event recorded on the same date as their date of detection of diabetes (data not shown).

Among clinically detected cases who were screening participants, those who had a previous diabetic or non-diabetic hyperglycaemic screening result were diagnosed on average 6.3 and 6.9 years, respectively, after their last screening. Cases who had previously had a normoglycaemic screening result were diagnosed on average 10.3 years after their last screening; among cases who had screened negative for diabetes or non-diabetic hyperglycaemia, n=13 (0.76%) were diagnosed with diabetes within 1 year (data not shown). Overall, among screening participants, screen-detected cases had similar mean levels of self-reported bad health and blood pressure to clinically detected cases who had had a diabetic or non-diabetic hyperglycaemic OGTT result at previous screening (Table 2). For total serum cholesterol the average levels were also very similar (Δ ≤ 0.3 mmol/L) although the differences were statistically significant. Compared to screen-detected cases, mean BMI was 1.1 kg/m² lower among clinically detected cases who were screening participants. Among screen-detected cases, n=443 (43.3%) reported family history of diabetes in the VIP questionnaire; the corresponding number was n=1,519 (33.7%) among clinically detected cases who were screening participants (data not shown).

Table 2.

Characteristics measured at concurrent or previous screening among type 2 diabetes mellitus cases who were screening participants, Västerbotten Intervention Programme 1992-2013.

						Incident clinically detected cases, screening participants (n=4,561)
	Confirmed Screen-detected cases (n=1,024)		Unconfirmed screen-positive cases (n=1,403)			All (n=4,506)			Previous diabetic OGTT > 1 year pre-detection (n=776)			Previous non-diabetic hyperglycaemia (n=1,898)			Previous normoglycaemia (n=1,716)
	mean	SD	mean	SD	p-value	mean	SD	p-value	mean	SD	P-value	mean	SD	p-value	mean	SD	p-value
Age at concurrent or previous screening (years)	55.1	6.4	53.4	8.0	-	53.4	7.9	-	54.2	7.3	-	54.3	7.2	-	51.9	8.7	-
Time from screening to first register detection (years)	0.1	0.2	-		-	8.1	4.9	-	6.2	4.1	-	6.9	4.5	-	10.3	4.9	-
BMI (kg/m2)	30.8	5.6	27.8	4.8	<0.001	29.7	4.7	<0.001	30.3	5.0	0.048	29.9	4.7	<0.001	29.1	4.4	<0.001
Self-reported overall bad health (n, %)	397	38.8	459	32.7	0.002	1,795	39.8	0.741	323	41.6	0.455	757	39.9	0.588	662	38.6	0.993
Systolic blood pressure (mmHg)	140.9	18.6	136.6	20.5	<0.001	139.2	18.7	0.010	140.8	19.2	0.967	140.3	18.6	0.394	137.0	18.4	<0.001
Diastolic blood pressure (mmHg)	86.2	10.6	83.2	11.5	<0.001	85.2	11.1	0.007	86.0	10.8	0.638	85.4	10.6	0.051	84.3	11.3	<0.001
Total serum cholesterol (mmol/L)	5.6	1.1	5.5	1.2	0.011	5.8	1.2	<0.001	5.8	1.2	0.002	5.7	1.2	0.009	5.8	1.2	<0.001

Results not included for n=116 cases who were clinically detected screening participants but had missing information on OGTT results (glycaemic status) from previous screening.

Unconfirmed screen-positive cases had only a diabetic screening result, whereas confirmed screen-detected cases had a diabetic screening result and a medical or prescription record of diabetes within one year. Clinically detected diabetes cases were identified in five sources of medical and prescription records, unrelated to screening.

p-values are for comparisons vs. screen-detected cases.

Missing data: BMI, n=49; Self-reported health, n=78; systolic blood pressure, n=82; diastolic blood pressure, n=83; Total serum cholesterol, n=61.

BMI: Body mass index, OGTT: Oral glucose tolerance test, SD: Standard deviation.

The StdMR was 3.2/1,000 PYR for all VIP participants, 8.4/1,000 PYR for all non-participants, and 3.0/1,000 PYR for known normoglycaemic VIP participants (data not shown). Average follow-up time was 8.7 (median 7.8) years for screen-detected diabetes cases and 7.2 (median 6.2) years for clinically detected cases after their date of detection(maximum 21.9 years). Screen-detected cases had a consistently lower rate of all-cause mortality, CVD, renal disease and retinopathy than clinically detected cases (Table 3). Among clinically detected cases, screening participants had lower rates of all outcomes compared to screening non-participants. There was a clear pattern of relative risks; compared to screen-detected diabetes cases, clinically detected diabetes cases who were screening participants had an increased risk (e.g. all-cause mortality HR 1.70, 95% CI 1.32, 2.18) and clinically detected cases who were screening non-participants had an even higher risk (e.g. all-cause mortality HR 2.31, 95% CI 1.82, 2.94) (Table 4). Results were similar over the course of follow-up; 3,397 (30.7%) diabetes cases, including 394 screen-detected cases, were followed 10 years or longer; the HR for all-cause mortality after 10 years was 1.91 (95% CI 1.27, 2.85) for clinically detected cases vs. screen-detected cases.

Table 3.

Crude and standardised incident event and mortality rates among type 2 diabetes mellitus cases, Västerbotten Intervention Programme 1992-2013.

	Deaths			CVD events			Renal Disease			Retinopathy
	n	MR	StdMR	n	IR	StdIR	n	IR	StdIR	n	IR	StdIR
Confirmed screen-detected cases	73	8.2	4.2	128	15.5	8.7	39	4.4	3.3	70	8.1	6.3
Unconfirmed screen-positive cases	139	10.4	5.9	141	11.0	5.8	23	1.7	0.9	9	0.7	0.4
Incident clinically detected cases	1330	21.4	15.5	1,704	30.5	21.9	649	10.8	9.3	757	12.7	12.7
Screening-participants	515	18.8	11.5	680	27.1	19.8	258	9.6	8.6	279	10.5	9.7
Previous diabetic OGTT > 1 year pre-detection	93	17.0	9.4	114	22.9	13.7	41	7.7	5.0	70	13.3	9.9
Previous non-diabetic hyperglycaemia	194	16.4	12.0	267	24.5	21.0	100	8.7	10.5	103	9.0	10.3
Previous normoglycaemia	206	22.0	12.5	275	32.3	20.4	102	11.2	7.9	91	10.0	8.8
Screening non-participants	815	23.4	18.4	1,024	33.3	25.5	391	11.6	10.2	478	14.6	14.2

IR and MR reported per 1,000 person-years. Age and sex standardised MR and IR are calculated with the total study population as reference.

Rates do not include events that coincide with date of diabetes detection.

CVD, renal and retinal events detected in Hospital Outpatient and Inpatient Discharge Registers, and the Cause of Death Register.

Unconfirmed screen-positive cases had only a diabetic screening result, whereas confirmed screen-detected cases had a diabetic screening result and a medical or prescription record of diabetes within one year. Clinically detected diabetes cases were identified in five sources of medical and prescription records, unrelated to screening.

CVD: Cardiovascular disease, IR: Incidence rate, MR: Mortality rate, OGTT: Oral glucose tolerance test, StdIR; Standardised Incidence Rate, StdMR: Standardised Mortality Rate.

Table 4.

Associations between mode of detection of type 2 diabetes mellitus and death, incident CVD events, renal disease, or retinopathy. Västerbotten Intervention Programme 1992-2013.

	All-cause mortality		CVD events		Renal Disease		Retinopathy
	HR	95% CI	HR	95% CI	HR	95% CI	HR	95% CI
Confirmed screen-detected cases	1	Ref	1	Ref	1	Ref	1	Ref
Unconfirmed screen-positive cases	1.35	1.01, 1.79	0.77	0.60, 0.98	0.41	0.25, 0.69	0.08	0.04, 0.16
Incident clinically detected cases	2.07	1.63, 2.62	1.55	1.29, 1.86	2.26	1.64, 3.13	1.66	1.30, 2.13
Screening-participants	1.70	1.32, 2.18	1.25	1.03, 1.52	1.89	1.34, 2.66	1.38	1.06, 1.80
Previous diabetic OGTT > 1 year pre-detection	1.61	1.18, 2.20	1.11	0.86, 1.43	1.58	1.02, 2.45	1.73	1.24, 2.42
Previous non-diabetic hyperglycaemia	1.59	1.21, 2.09	1.18	0.95, 1.46	1.76	1.21, 2.56	1.19	0.87, 1.61
Previous normoglycaemia	2.05	1.56, 2.70	1.47	1.19, 1.83	2.20	1.51, 3.20	1.35	0.98, 1.85
Screening non-participants	2.31	1.82, 2.94	1.77	1.47, 2.13	2.54	1.82, 3.53	1.85	1.44, 2.38

HRs adjusted for calendar time in time-scale and the co-variates age at diabetes detection, calendar year of diabetes detection, sex, and socioeconomic status reported in the 1990 census.

Socioeconomic status imputed for n=1,561 individuals who had missing information.

Unconfirmed screen-positive cases had only a diabetic screening result, whereas confirmed screen-detected cases had a diabetic screening result and a medical or prescription record of diabetes within one year. Clinically detected diabetes cases were identified in five sources of medical and prescription records, unrelated to screening.

CI: Confidence interval, CVD: Cardiovascular disease, HR: Hazard ratio, OGTT: Oral glucose tolerance test.

Among clinically detected cases who were screening participants, those who had had previous normoglycaemia tended to have higher rates of outcomes, with the exception of retinopathy, compared to those who had had previous diabetic OGTTs or non-diabetic hyperglycaemia (Table 3). Generally those who had had a previous diabetic OGTT and those who had had previous non-diabetic hyperglycaemia had very similar relative risks (e.g. all-cause mortality HR 1.61 and 1.59, respectively), whereas those who had had previous normoglycaemia had higher risks for all outcomes (e.g. all-cause mortality HR 2.05, 95% CI 1.56, 2.70) with the exception of retinopathy (Table 4).

To explore whether some of the effect of mode of detection among cases who participated in screening could be explained by differences in the cases’ health status, we conducted sensitivity analyses adjusting for presence of a prior CVD event, and additionally for several biomarkers measured at previous screening (listed in Table 2) as well as time since previous screening (ESM TTable 2). As a result when adjusting for prior CVD event status all estimates were attenuated but, with the exception of relative risk of CVD events (HR 1.16, 95% CI 0.95, 1.41), remained significant. Estimates were further attenuated when adjusting for additional biomarkers.

Unconfirmed screen-positive cases were on average 1.7 years younger than screen-detected cases at the date of their diabetic OGTT screening result (Table 1); they had a higher MR than screen-detected cases, but lower IR of all other outcomes and very low IR of renal disease and retinopathy (Table 3). Compared to confirmed screen-detected diabetes cases, unconfirmed screen-positive cases had a higher risk of all-cause mortality (HR 1.35, 95% CI 1.01, 1.79) but lower risk of CVD (HR 0.77, 95% CI 0.60, 0.98) and substantially lower relative risks of renal disease (HR 0.41, 95% CI 0.25, 0.69) and retinopathy (HR 0.08, 95% CI 0.04, 0.16) (Table 4).

Discussion

In this study of a population included in an organised universal screening programme for diabetes we found that a diagnosis of diabetes can be brought forward by an average of 4.6 years by screening asymptomatic individuals, and that screen-detected cases appear to fare better than clinically detected cases after their diagnosis.

The lead time is somewhat longer than the 3.3 years and 2.2 years estimated in previous studies [5, 7]. There are important differences with regards to screening interval and analytical approach between this study and the previous studies that may explain this difference; the Ely cohort allocated individuals to a screening (in 5-year intervals) and non-screening arm and lead time was calculated as the difference in median diabetes duration in the two study arms for screen and clinically detected cases, screening participants and non-participants combined. In the ADDITION-Denmark study high risk individuals were invited to screening at one time-point and lead time was calculated as the difference in median diabetes duration between screen vs. clinically detected cases in the whole group eligible to be screened. In this study we compared age at detection in screen-detected vs. clinically detected cases that had been eligible to be screened in 10-year intervals.

We found that screen-detected diabetes cases had lower rates of all-cause mortality and incident CVD, renal disease and retinopathy than clinically detected cases. This is in line with the modelled estimated reduction in CVD events due to earlier routine treatment found in a previous study [24]. It is possible that the observed effect may be caused by the treatment that screen-detected cases presumably received earlier than clinically detected cases, but there are three important biases which may explain some of the effect.

• 1)Healthy user bias: Clinically detected cases who were screening non-participants were detected on average 3.6 years earlier than clinically detected cases who were screening participants, but despite being diagnosed earlier they had consistently worse health outcomes. On average VIP non-participants had more than twice as high all-cause mortality that VIP participants when comparing age and sex standardised mortality rates. Similarly, it has been shown that in screening for human papilloma virus, regular non-attenders have about a two-fold higher all-cause mortality than regular attenders [25]. Although VIP is not a screening programme for diabetes, it is likely that the individual choice to attend the clinical examinations would be guided by similar behaviour to the choice to attend systematic organised screening programmes. In VIP, participation has been linked to marital status and higher income, but not education [12].
• 2)Length time bias: The idea that slowly developing disease with a longer asymptomatic pre-clinical screen-detectable course is also more likely to have a long clinical course and better prognosis [26], has not been explored specifically for diabetes previously. However, our data indicate that this concept may be equally important for diabetes screening as it is for several cancers [27]. It appears that slowly progressing hyperglycaemia and diabetes may be associated with better health outcomes compared to more rapidly progressing disease, as indicated by the fact that clinically detected cases who had been normoglycaemic at their previous VIP examination had worse health outcomes after diagnosis compared to those who had been non-diabetic hyperglycaemic or who had had diabetic OGTT results. However, those who had diabetic or non-diabetic hyperglycaemia at their previous screening should have received lifestyle advice and referrals to continued care which could have also contributed to a better prognosis after diagnosis. In addition, the mean time to diabetes detection from screening was about 4 years longer for those with previous normoglycaemia and we cannot know for how long they would have lived with hyperglycaemia prior to their diagnosis. Ideally we would have liked to test the contribution of length time bias by adjusting for health status and biomarkers for diabetes severity at the time of detection, but these data were not available. When we adjusted for several biomarkers associated with general health status measured at previous screening the estimates were attenuated which supports a role of length time bias, but the analysis has limitations so a cautious interpretation is warranted.
• 3)Lead time bias: Although we adjust for age at detection in the analyses, we cannot disregard that there may be residual bias from differences in lead time as screen-detected cases had on average 1.5 years longer observation time due to being detected earlier in the disease course than clinically detected cases [28, 29].

Taken together, these data suggest that there may be a positive effect of early detection and treatment due to screening on survival and health outcomes after a diabetes diagnosis, but how much is not within the scope of this study to determine. These results are in line with those from ADDITION-Denmark [7].

We found that it was more common for a diabetic OGTT result to remain unconfirmed than to be confirmed within one year, which is in line with findings from previous studies [8]. In this study unconfirmed screen-positive cases were overall younger and had consistently better health at the point of the positive diabetic screening result than screen-detected confirmed cases. They also had lower IR of CVD and renal disease, considerably lower IR of retinopathy, but a higher MR. There is reason to believe that some of the difference in retinopathy and renal disease rates is due to surveillance bias as confirmed diabetes patients are more likely to be tested for these conditions, but this is less likely to be the case for CVD events and is not the case for deaths. These results indicate that unconfirmed screen-positive cases would potentially benefit from treatment of glycaemia and related risk factors in order to reduce risk of CVD [30].

The primary strength of this study was the large population size and that we were able to study a model for an organised whole population-based screening programme for diabetes with follow-up for over 20 years that included participants as young as age 30 years. The main limitations were relatively short follow-up period and that we could not assess the association between screening in general and health outcomes after diagnosis due to the non-randomised design (no non-screening control group). The diagnostic criteria for diabetes were revised during the study period when the OGTT fasting glucose level threshold was lowered from 7.8 mmol/L to 7.0 mmol/L in 1999 [31], although it was unclear when this revision was implemented in VIP, meaning that some cases may have been misclassified. However, there were only 10 screen-detected cases (data not shown) within this range between 1992 and 1998 and since all cases in this group were confirmed by another source within one year the resulting bias is likely limited. The median follow-up time was relatively short at 6.2-7.8 years but the results were similar even after 10 years follow-up. We did not have access to data on marital status and income, variables which have been associated with propensity to participate in screening [12], but we were able to control for SES. Clinically detected cases were identified from 5 different sources and as a consequence systematic information on biomarkers associated with severity of diabetes at time of diagnosis was unavailable.

In conclusion, in this population-based study of screen and clinically detected diabetes cases we found that screen-detected cases were detected at a younger age, and may have better survival and lower rates of CVD, renal disease and retinopathy than clinically detected cases.

Abbreviations

CVD

Cardiovascular disease

SES

Socioeconomic status

VIP

Västerbotten Intervention Programme