Study population, recorded DR prevalence and incidence

Revision 03: ใช้เฉพาะผู้ที่ข้อมูลครบในทุกตัวแปรของ Table 1; สร้างโรคร่วมใหม่จาก OPD/IPD โดยรวมประวัติหลังปี 2018 แล้ว ใช้สี่ภาค และคำนวณ annual prevalence พร้อม 95% CI ใหม่หลังคัดข้อมูลขาด

Study population

Before complete-case selection, 5,022,524 unique people met the prevalence eligibility criteria during 2017 to 30 September 2025 and 4,205,887 met at least one annual incidence risk-set criterion during 2019 to 30 September 2025. After excluding people missing any Table 1 characteristic at their first eligible entry into the respective analysis, the populations comprised 4,676,861 and 3,977,599 people, respectively. Among the retained prevalence population, 349,604 (7.48%) had H36-defined DR status in at least one eligible year. Table 1 describes the two populations separately. Recorded DR status was carried forward for annual prevalence. The incidence analysis used rolling two-year H36-free look-back windows and established recorded diabetes before each window; it is a separate analysis population. Comorbidity history was rebuilt from available OPD/IPD records from 2015 until the day before each person's entry. No qualifying code in that look-back was classified as No. Annual prevalence is reported with pointwise 95% Wilson confidence intervals. The 2025 results cover January to September only.

Table 1. Characteristics of the prevalence and incidence analysis populations

CharacteristicPrevalence, 2017-Sep 2025
N=4,676,861
Incidence, 2019-Sep 2025
N=3,977,599
Analysis population
Unique people4,676,861 (100.00)3,977,599 (100.00)
Sex
Male1,840,547 (39.35)1,522,587 (38.28)
Female2,836,314 (60.65)2,455,012 (61.72)
Age at cohort entry, years
<1817,065 (0.36)10,532 (0.26)
18-39319,538 (6.83)187,591 (4.72)
40-49766,275 (16.38)502,587 (12.64)
50-591,452,787 (31.06)1,128,810 (28.38)
60-691,300,549 (27.81)1,221,407 (30.71)
70-79612,996 (13.11)659,828 (16.59)
80+207,651 (4.44)266,844 (6.71)
Geographic region of linked facility
North947,306 (20.26)811,123 (20.39)
Northeast1,736,504 (37.13)1,476,771 (37.13)
Central1,394,478 (29.82)1,181,233 (29.70)
South598,573 (12.80)508,472 (12.78)
Facility type
Regional / general hospital1,066,746 (22.81)880,802 (22.14)
Community hospital1,755,766 (37.54)1,451,215 (36.48)
Primary / community health facility1,615,156 (34.54)1,455,161 (36.58)
Other public / non-MOPH hospital178,702 (3.82)140,729 (3.54)
Private hospital / clinic59,562 (1.27)48,866 (1.23)
Other facility929 (0.02)826 (0.02)
Time since first recorded diabetes at entry
<1 year2,291,373 (48.99)0 (0.00)
1-<5 years1,052,785 (22.51)2,131,165 (53.58)
5-<10 years893,077 (19.10)1,148,082 (28.86)
10+ years439,626 (9.40)698,352 (17.56)
Hypertension
Yes1,933,087 (41.33)2,298,478 (57.79)
No2,743,774 (58.67)1,679,121 (42.21)
Dyslipidaemia
Yes1,278,714 (27.34)1,872,880 (47.09)
No3,398,147 (72.66)2,104,719 (52.91)
CKD / recorded diabetic renal complications
Yes440,057 (9.41)791,991 (19.91)
No4,236,804 (90.59)3,185,608 (80.09)
Ischaemic heart disease
Yes117,071 (2.50)166,113 (4.18)
No4,559,790 (97.50)3,811,486 (95.82)
Stroke / specified sequelae
Yes138,687 (2.97)202,454 (5.09)
No4,538,174 (97.03)3,775,145 (94.91)
Heart failure (I50)
Yes65,869 (1.41)114,719 (2.88)
No4,610,992 (98.59)3,862,880 (97.12)

Values are n (column %). Each person is counted once within each analysis population. Age and recorded diabetes duration are assessed at first eligible entry into the respective analysis. People missing any reported characteristic are excluded from that analysis population; age below 18 is retained when known. The two populations overlap but need not be nested because their entry dates and completeness assessments differ.

Comorbidities use qualifying diagnosis codes recorded from 1 January 2015 until the day before individual entry. A person entering on 1 January 2019 therefore has the 2015-2018 look-back. Look-back length varies with entry date, so these proportions should not be read as directly comparable clinical risks between the two populations. No qualifying code is coded No under the operational study definition; it does not prove clinical absence. All extended 2019-Sep2025 extracts must complete before absence is coded No. The renal category includes recorded diabetic renal complications and is not an eGFR-confirmed CKD classification.

Region is based on the province of the linked first-recorded-diabetes facility, not residence. Study groupings comprise North (17 provinces), Northeast (20), Central (26, including Bangkok and eastern/western provinces), and South (14); the province crosswalk is retained. Facility categories reflect the supplied reference snapshot.

Table 2. Annual recorded DR prevalence and 95% confidence intervals

YearEligible DMRecorded DR statusPrevalence (%)95% CI (%)
20172,922,45587,5262.992.98-3.01
20183,126,980129,1104.134.11-4.15
20193,290,350161,6114.914.89-4.94
20203,439,712185,0365.385.36-5.40
20213,579,151200,6125.615.58-5.63
20223,730,059221,0165.935.90-5.95
20233,904,513249,6436.396.37-6.42
20244,053,798280,2276.916.89-6.94
2025*4,151,969303,0217.307.27-7.32

DR is H36 in any diagnosis position. Recorded status is carried forward from the first H36 within the 2017-Sep2025 study window. Annual denominators comprise retained complete cases with diabetes recorded by period end and no recorded death by that end. Prior H36 before recorded DM is not automatically excluded. * 2025 ends 30 September. Eligibility uses recorded death because disenrollment dates are unavailable.

Figure 1. Annual recorded DR prevalence

Connected annual prevalence estimates with 95% Wilson confidence intervals

Pointwise 95% Wilson binomial intervals describe model-based precision of each recorded proportion. They do not cover under-recording, incomplete historical coverage, selection from complete-case exclusions, or repeated-person correlation across years. The dashed final segment identifies the shorter 2025 observation window.

Figure 2. Recorded DR prevalence by demographic characteristics

Recorded DR prevalence by demographic characteristics: crude period prevalence with 95% Wilson intervals

Crude period recorded DR prevalence, 2017-30 September 2025, in the same complete-case prevalence population as Table 1. Each category uses its own DM denominator; H36 status and eligibility follow the period-prevalence definition. The dashed vertical line marks overall prevalence. Characteristics are assessed at first eligible entry. These proportions are unadjusted for other characteristics and observation length; the intervals are pointwise 95% Wilson CIs.

Figure 3. Recorded DR prevalence by baseline comorbidity

Recorded DR prevalence by baseline comorbidity: crude period prevalence with 95% Wilson intervals

Crude period recorded DR prevalence, 2017-30 September 2025, in the same complete-case prevalence population as Table 1. Each category uses its own DM denominator; H36 status and eligibility follow the period-prevalence definition. The dashed vertical line marks overall prevalence. Characteristics are assessed at first eligible entry. These proportions are unadjusted for other characteristics and observation length; the intervals are pointwise 95% Wilson CIs.

Figure 4. Recorded DR prevalence across four geographic regions

Four-region DR prevalence map with 95% Wilson confidence intervals; unrepresented areas are grey

The map uses the same crude period prevalence estimates for 2017-30 September 2025 as the regional forest plot. Province boundaries are shown for orientation; colours represent regional estimates based on linked facility province, not residence. Areas without represented patients are grey and are not assigned zero prevalence. Only four regional estimates and the overall estimate are reported; there is no separate Bangkok estimate.

Table 3. Sensitivity analysis of overall recorded prevalence

DR status carry-forward
rule
DR cases (n)DM population (n)Prevalence (%)
Primary: until disenrollmentxxxxxxxxx
1 yearxxxxxxxxx
2 yearsxxxxxxxxx
3 yearsxxxxxxxxx
4 yearsxxxxxxxxx
5 yearsxxxxxxxxx

หมายเหตุ: ตารางนี้เป็นโครงร่างและยังไม่ได้วิเคราะห์ ไม่แนะนำให้วิเคราะห์ในรูปแบบนี้โดยตรง เนื่องจากการไม่มี H36 ซ้ำอาจเกิดจากการไม่มาตรวจตาหรือไม่ได้ลงรหัสระหว่างรับบริการอื่น การกำหนดให้สถานะหมดอายุจึงอาจสะท้อนรูปแบบบริการและการลงรหัส มากกว่าสถานะ DR จริง

Table 4. Incidence of recorded diabetic retinopathy, 2019–2025

YearEligible population
at start of year, n
Recorded DR
events, n
Person-years Incidence proportion,
% (95% CI)
Incidence rate per 1,000
person-years (95% CI)
20192,505,78330,1332,453,377.21.20 (1.19–1.22)12.28 (12.14–12.42)
20202,702,91225,0942,663,975.30.93 (0.92–0.94)9.42 (9.30–9.54)
20212,895,04121,3482,846,619.40.74 (0.73–0.75)7.50 (7.40–7.60)
20223,062,55027,6613,007,022.50.90 (0.89–0.91)9.20 (9.09–9.31)
20233,195,23436,2603,136,003.21.13 (1.12–1.15)11.56 (11.44–11.68)
20243,316,63137,6083,268,203.51.13 (1.12–1.15)11.51 (11.39–11.62)
2025*3,464,77125,5082,569,762.30.74 (0.73–0.75)9.93 (9.80–10.05)

Eligibility is assessed at 1 January of each year: recorded diabetes before the two-calendar-year look-back window, no recorded death before year start, and no H36 in that window, within the retained complete-case incidence population. Events are the first H36 in the index year after this code-free window, in any diagnosis position. Earlier H36 outside the window does not exclude re-entry; events therefore need not represent first-ever DR or biological onset. Each person contributes at most one event per year, and people may appear in multiple years; annual denominators must not be summed as unique people.

Incidence proportion = events / eligible population at year start × 100, with two-sided 95% Wilson score intervals; this is a crude observed proportion, not a survival-based risk estimate. Incidence rate = events / person-years × 1,000, with two-sided exact Poisson (Garwood) intervals. Follow-up runs from 1 January to the earliest qualifying H36, recorded death, or period end; elapsed days are divided by 365.25. Death on the event date is treated as an event. Disenrollment dates are unavailable; recorded history does not establish continuous observation.

* 2025 covers 1 January–30 September only. Its proportion is not directly comparable with a full-year proportion. Person-time adjusts for differing follow-up length but not seasonality, service use or incomplete recording. Intervals are pointwise model-based precision estimates and do not account for coding/selection bias or dependence between annual estimates. No formal trend test is presented.

Figure 5. Recorded DR incidence proportion and incidence rate

Two panels of annual recorded DR incidence proportion and rate with 95% confidence intervals; 2025 is January to September only

The dashed final segment and shaded area identify partial-year 2025. The two panels use separate units and vertical scales; confidence intervals may appear narrow because the denominators are large.

Factors associated with incident recorded DR: Fine–Gray competing-risk regression

The analysis estimates associations between baseline characteristics and the cumulative incidence of first recorded DR after cohort entry, with death before DR as a competing event. It includes 3,846,216 people, 169,282 DR events, 391,162 competing deaths, and 3,285,772 administrative censored observations. Observed follow-up to first event or administrative end totals 19,409,385.3 person-years. This cohort is distinct from the annual rolling-window incidence risk sets in Table 4.

ElementOperational definition and analysis principle
Entry and populationUse the retained complete-case incidence population. Each person enters once, at the first eligible 1 January in 2019–2025, with recorded DM before the start of the preceding two full calendar years. Entry is time zero. Eligibility uses information available through entry; persons do not re-enter after an event.
Prior DR exclusionExclude H36 in any diagnosis position anywhere in the available history through and including entry day, even if older than two years. All available OPD/IPD H36 sources are searched; the earliest observed record is 16 October 2014, without assuming complete uniform coverage back to that date. Excluded: 124,522 people with prior/same-day H36 and another 6,861 with recorded death on/before entry.
DM durationBaseline recorded DM duration = (entry date − first recorded DM diagnosis date)/365.25. The DM date is the earliest parseable CHRONIC E10–E14 date_diag, not verified biological onset. Adjust using 1–<5, 5–<10 and ≥10 years; 1–<5 is reference. The history requirement leaves no <1-year group.
Event codingStatus 1: first H36 strictly after entry, no later than recorded death and no later than 30 September 2025. Status 2: death before qualifying DR. Status 0: administrative end without either event. One DR event maximum per person. This is first recorded DR among those without a detected prior code, not proven first-ever disease onset.
Dates and observationObserved follow-up ends at the earliest DR, death or 30 September 2025. Elapsed days are used in computation; years = days/365.25. Same-day DR/death is assigned DR (9 people). Year–month-only death dates follow the existing first-day-of-month convention, so ordering is approximate. Disenrollment and validated loss-to-follow-up dates are unavailable.
Baseline factorsMutually adjust sex, age group, linked-facility region/type, recorded DM duration, hypertension, dyslipidaemia, CKD/recorded diabetic renal complications, ischaemic heart disease, stroke/specified sequelae and heart failure. Factors are fixed at entry. Comorbidities use qualifying codes from 1 January 2015 strictly before entry; no qualifying code = No under the operational definition. Region is facility province, not residence.
Fine–Gray modelFit a proportional subdistribution-hazards regression for DR with death as the competing event. Estimate the censoring distribution separately within calendar-entry-year groups and stratify the baseline subdistribution hazard by entry year. Use Breslow handling of tied event days and subject-cluster robust sandwich standard errors; report adjusted subdistribution HRs (sHRs) and robust Wald 95% CIs.
Competing-death risk setsSubjects who die before DR are retained in the mathematical Fine–Gray risk sets according to the model's weighting rule; they are not treated as people who can physically develop DR after death. Within each entry-year group, the only censoring is a common administrative horizon, so inverse-censoring weights are exactly one before that horizon. R finegray verifies this for every transformed row. The extended risk-set time is not reported as observed person-time.
InterpretationsHR relates baseline factors to the DR cumulative incidence function in the presence of competing death. It is not an ordinary hazard ratio among survivors, a cumulative-risk ratio, or proof of a causal effect. Associations can reflect both DR recording and competing mortality. A lower estimate in an older or severely ill group does not establish protection from DR.
Proportionality diagnosticUse a prespecified robust efficient-score test for factor × log(1 + follow-up years), projecting out the baseline coefficients. The subject-score sandwich and information matrices use the full Fine–Gray risk sets. The calculation is independently checked against R and an explicitly time-split reference model. This tests a particular time trend; it is not proof of all forms of proportionality when nonsignificant.
AJ estimatorRetain the unadjusted Aalen–Johansen cumulative incidence curves, with death before DR as a competing event. Update F_DR(t)=F_DR(t−)+S(t−)d_DR(t)/Y(t), with S based on both first-event types. Use the common pre-event risk set for ties and censor after events. Pointwise 95% log-log CIs use validated multinomial transition-increment covariance propagation. AJ is descriptive; Fine–Gray provides covariate-adjusted associations.

Historical code absence is not a negative retinal examination or proof of continuous enrollment. Source coverage, service use, coding, selection, referral patterns and approximate death dates affect interpretation; confidence intervals do not cover those biases. AJ curves pool entry cohorts with different administrative follow-up lengths and assume noninformative censoring within reported groups. Table 1 describes the candidate incidence population, not the final Fine–Gray cohort.

Time-varying associations: the Fine–Gray log-time diagnostic rejects proportional subdistribution hazards globally (χ²=6,387.25, df=23, p<0.001). The overall sHRs in Table 5 are follow-up summaries, not constant effects. Table 6 and Figure 6 provide interval-specific subdistribution-HR summaries for interpretation.

Table 5. Mutually adjusted Fine–Gray subdistribution hazard ratios

CharacteristicPeople, nDR events, nAdjusted sHR (95% CI)p-value
Sex
Male1,478,56456,2101.00 (Reference)—
Female2,367,652113,0721.11 (1.10–1.12)<0.001
Age at cohort entry (years)
<1810,4332430.47 (0.42–0.54)<0.001
18-39184,2279,3161.00 (Reference)—
40-49490,56628,8430.97 (0.94–0.99)0.005
50-591,092,43961,8240.78 (0.76–0.80)<0.001
60-691,173,91950,3700.49 (0.48–0.50)<0.001
70-79635,25516,2050.25 (0.24–0.26)<0.001
80+259,3772,4810.09 (0.09–0.10)<0.001
Region of linked facility
North786,66032,4361.00 (Reference)—
Northeast1,429,34759,1141.04 (1.02–1.05)<0.001
Central1,137,34251,8011.24 (1.22–1.26)<0.001
South492,86725,9311.35 (1.33–1.38)<0.001
Type of linked facility
Regional / general hospital845,07339,8691.00 (Reference)—
Community hospital1,406,05958,4790.77 (0.76–0.78)<0.001
Primary / community health facility1,408,34466,6440.83 (0.82–0.84)<0.001
Other public / non-MOPH hospital138,0302,4630.59 (0.57–0.62)<0.001
Private hospital / clinic47,9061,7600.84 (0.80–0.88)<0.001
Other facility804671.51 (1.19–1.92)<0.001
Recorded DM duration at entry
1-<5 years2,098,69859,5981.00 (Reference)—
5-<10 years1,105,89265,9311.51 (1.49–1.53)<0.001
10+ years641,62643,7531.90 (1.87–1.93)<0.001
Hypertension
No1,662,39654,4961.00 (Reference)—
Yes2,183,820114,7861.42 (1.40–1.44)<0.001
Dyslipidaemia
No2,073,28071,7171.00 (Reference)—
Yes1,772,93697,5651.42 (1.40–1.43)<0.001
CKD / recorded diabetic renal complications
No3,119,939126,2371.00 (Reference)—
Yes726,27743,0451.29 (1.28–1.31)<0.001
Ischaemic heart disease
No3,693,419162,1621.00 (Reference)—
Yes152,7977,1200.99 (0.96–1.02)0.412
Stroke / specified sequelae
No3,659,480161,7631.00 (Reference)—
Yes186,7367,5190.93 (0.91–0.96)<0.001
Heart failure (I50)
No3,745,793164,6821.00 (Reference)—
Yes100,4234,6001.00 (0.96–1.03)0.808

References: male; age 18–39; North; regional/general hospital; recorded DM duration 1–<5 years; No for each comorbidity. Age <18 is retained and listed first for ascending age order. All factors are adjusted simultaneously. Entry-year baseline strata have no reported coefficient. Counts refer to the full single-entry cohort. Large N makes small differences statistically detectable; sHR is not a risk ratio.

Table 6. Fine–Gray subdistribution HRs by follow-up period

Characteristic0–<2 years
sHR (95% CI)
2–<4 years
sHR (95% CI)
≥4 years
sHR (95% CI)
Sex
Male1.00 (Reference)1.00 (Reference)1.00 (Reference)
Female1.04 (1.03–1.06)1.09 (1.07–1.12)1.20 (1.18–1.23)
Age at cohort entry (years)
<180.50 (0.39–0.63)0.49 (0.39–0.63)0.43 (0.35–0.52)
18-391.00 (Reference)1.00 (Reference)1.00 (Reference)
40-491.04 (1.00–1.09)0.92 (0.88–0.96)0.94 (0.91–0.97)
50-590.92 (0.88–0.96)0.76 (0.73–0.80)0.69 (0.66–0.71)
60-690.68 (0.65–0.70)0.49 (0.47–0.51)0.37 (0.36–0.38)
70-790.39 (0.37–0.40)0.25 (0.24–0.26)0.16 (0.16–0.17)
80+0.17 (0.16–0.18)0.09 (0.08–0.09)0.04 (0.04–0.05)
Region of linked facility
North1.00 (Reference)1.00 (Reference)1.00 (Reference)
Northeast1.02 (0.99–1.04)0.99 (0.96–1.02)1.10 (1.07–1.12)
Central1.23 (1.20–1.26)1.17 (1.14–1.20)1.30 (1.27–1.33)
South1.35 (1.31–1.39)1.27 (1.22–1.31)1.43 (1.39–1.47)
Type of linked facility
Regional / general hospital1.00 (Reference)1.00 (Reference)1.00 (Reference)
Community hospital0.67 (0.66–0.69)0.74 (0.72–0.76)0.93 (0.91–0.95)
Primary / community health facility0.75 (0.74–0.77)0.80 (0.78–0.82)0.94 (0.92–0.96)
Other public / non-MOPH hospital0.57 (0.53–0.61)0.64 (0.59–0.69)0.59 (0.55–0.64)
Private hospital / clinic0.77 (0.71–0.83)0.87 (0.79–0.95)0.92 (0.85–1.00)
Other facility1.63 (1.14–2.33)1.74 (1.14–2.68)1.16 (0.72–1.88)
Recorded DM duration at entry
1-<5 years1.00 (Reference)1.00 (Reference)1.00 (Reference)
5-<10 years1.52 (1.49–1.56)1.56 (1.52–1.61)1.48 (1.45–1.51)
10+ years2.06 (2.01–2.11)1.97 (1.91–2.03)1.76 (1.72–1.80)
Hypertension
No1.00 (Reference)1.00 (Reference)1.00 (Reference)
Yes1.68 (1.64–1.71)1.38 (1.35–1.42)1.24 (1.22–1.27)
Dyslipidaemia
No1.00 (Reference)1.00 (Reference)1.00 (Reference)
Yes1.46 (1.43–1.49)1.39 (1.36–1.42)1.39 (1.37–1.42)
CKD / recorded diabetic renal complications
No1.00 (Reference)1.00 (Reference)1.00 (Reference)
Yes1.47 (1.44–1.50)1.22 (1.19–1.25)1.15 (1.13–1.18)
Ischaemic heart disease
No1.00 (Reference)1.00 (Reference)1.00 (Reference)
Yes1.04 (1.00–1.08)0.97 (0.92–1.02)0.91 (0.87–0.96)
Stroke / specified sequelae
No1.00 (Reference)1.00 (Reference)1.00 (Reference)
Yes1.08 (1.04–1.12)0.88 (0.84–0.93)0.74 (0.71–0.78)
Heart failure (I50)
No1.00 (Reference)1.00 (Reference)1.00 (Reference)
Yes1.27 (1.22–1.33)0.90 (0.84–0.96)0.63 (0.59–0.68)

Separate fits use the original Fine–Gray extended risk sets within approximate two-year bands (0–730, >730–1461 and >1461 elapsed days), with the same baseline factors, entry-year strata and subject-robust standard errors. Prior competing deaths remain in the extended risk sets; these are not landmark analyses restricted to survivors. The estimates permit differences between bands but remain summaries within bands; they do not establish proportionality within each band. No formal test comparing the separate period estimates is presented.

Figure 6. Adjusted Fine–Gray subdistribution hazard ratios

Adjusted Fine-Gray subdistribution hazard ratios with robust 95% confidence intervals

Figure 7. Overall AJ cumulative incidence with competing death

At five years, first recorded DR cumulative incidence is 3.93% (95% CI 3.91–3.95); death before recorded DR is 10.08% (95% CI 10.05–10.11). These are unadjusted absolute cumulative incidences; sHRs do not equal ratios of these values.

AJ cumulative incidence of first recorded DR and competing death

Figure 8. AJ cumulative incidence by baseline demographic characteristics

Unadjusted AJ DR cumulative incidence by demographic characteristics

Figure 9. AJ cumulative incidence by baseline clinical history

Unadjusted AJ DR cumulative incidence by six baseline comorbidities

Table 7. Unadjusted AJ cumulative incidence of recorded DR

Baseline characteristicPeople at entry, n3-year CIF, % (95% CI)5-year CIF, % (95% CI)
Overall
Overall3,846,2162.36 (2.34–2.38)3.93 (3.91–3.95)
Sex
Male1,478,5642.16 (2.14–2.19)3.53 (3.50–3.56)
Female2,367,6522.48 (2.46–2.50)4.17 (4.14–4.19)
Age at cohort entry (years)
<1810,4331.07 (0.88–1.30)2.07 (1.78–2.39)
18-39184,2272.44 (2.37–2.52)4.85 (4.74–4.97)
40-49490,5662.87 (2.82–2.92)5.32 (5.25–5.40)
50-591,092,4392.92 (2.89–2.95)5.08 (5.04–5.13)
60-691,173,9192.47 (2.44–2.50)3.89 (3.85–3.93)
70-79635,2551.57 (1.54–1.60)2.28 (2.24–2.32)
80+259,3770.67 (0.63–0.70)0.89 (0.85–0.92)
Region of linked facility
North786,6602.24 (2.21–2.27)3.66 (3.62–3.70)
Northeast1,429,3472.19 (2.16–2.21)3.61 (3.58–3.65)
Central1,137,3422.47 (2.44–2.50)4.17 (4.13–4.21)
South492,8672.81 (2.76–2.85)4.72 (4.65–4.78)
Type of linked facility
Regional / general hospital845,0732.73 (2.70–2.77)4.39 (4.34–4.44)
Community hospital1,406,0592.18 (2.15–2.20)3.74 (3.71–3.78)
Primary / community health facility1,408,3442.44 (2.41–2.46)4.01 (3.98–4.05)
Other public / non-MOPH hospital138,0301.09 (1.04–1.15)1.84 (1.76–1.93)
Private hospital / clinic47,9062.31 (2.17–2.46)3.85 (3.66–4.05)
Other facility8045.43 (3.98–7.18)8.39 (6.51–10.57)
Recorded DM duration at entry
1-<5 years2,098,6981.81 (1.79–1.83)3.06 (3.03–3.08)
5-<10 years1,105,8922.70 (2.67–2.73)4.43 (4.39–4.47)
10+ years641,6263.41 (3.36–3.45)5.27 (5.22–5.33)
Hypertension
No1,662,3961.54 (1.52–1.56)2.85 (2.82–2.87)
Yes2,183,8202.98 (2.96–3.00)4.74 (4.71–4.77)
Dyslipidaemia
No2,073,2801.70 (1.68–1.72)2.98 (2.95–3.00)
Yes1,772,9363.15 (3.12–3.17)5.07 (5.03–5.10)
CKD / recorded diabetic renal complications
No3,119,9392.10 (2.09–2.12)3.63 (3.60–3.65)
Yes726,2773.45 (3.40–3.49)5.16 (5.11–5.22)
Ischaemic heart disease
No3,693,4192.33 (2.31–2.35)3.90 (3.88–3.93)
Yes152,7973.11 (3.02–3.20)4.46 (4.35–4.57)
Stroke / specified sequelae
No3,659,4802.34 (2.32–2.35)3.92 (3.90–3.94)
Yes186,7362.84 (2.76–2.92)4.08 (3.98–4.18)
Heart failure (I50)
No3,745,7932.33 (2.31–2.35)3.91 (3.89–3.93)
Yes100,4233.54 (3.42–3.66)4.63 (4.49–4.77)

Pointwise CIs are not simultaneous bands. Longer-horizon estimates rely on earlier-entry cohorts still observed. The existing individual AJ graphs include numbers-at-risk tables; all AJ estimates and figures are unchanged by replacing the regression model.

Methods: Fine–Gray transformation and subject-robust inference; AJ and competing risks. Fine–Gray models were fitted using R survival 3.8.3; numerical checks independently reproduce the full-data likelihood, scores and robust covariance.