• Open Access

Does higher quality of primary healthcare reduce hospital admissions for diabetes complications? A national observational study

Authors


Abstract

Aim

To determine if hospital admission rates for diabetes complications (acute complications, chronic complications, no complications and hypoglycaemia) were associated with primary care diabetes management.

Methods

We performed an observational study in the population in England during the period 2004–2009 (54 741 278 people registered with 8140 general practices). We used multivariable negative binomial regression to model the associations between indirectly standardized hospital admission rates for complications and primary healthcare quality, supply and access indicators, diabetes prevalence and population factors.

Results

In multivariate regression models, increasing deprivation (incidence rate ratio: 1.0154; < 0.001, 95% CI 1.0141–1.0166) and diabetes prevalence (incidence rate ratio: 1.0956; < 0.001, 95% CI 1.0677–1.1241) were risk factors for admission, while most healthcare covariates, i.e. a larger practice population (incidence rate ratio 0.9999, = 0.013, 95% CI 0.9999–0.9999), better patient-perceived urgent and non-urgent access to primary care (incidence rate ratio: 0.9989, = 0.023; 95% CI 0.9979–0.9998 and incidence rate ratio: 0.9988; = 0.003, 95% CI 0.9980–0.9996, respectively) and better HbA1c target achievement (incidence rate ratio: 0.9971; < 0.001, 95% CI 0.9958–0.9984), were protective. Diabetes admissions decreased significantly during the period 2004–2009.

Conclusions

After controlling for population factors, better scheduled primary care access and glycaemic control were associated with lower hospital admission rates across most complications. There is little rationale to restrict primary care-sensitive condition definitions to acute complications. They should be revised to improve the usefulness of hospital admission data as an outcome measure, and to facilitate international comparisons. The risk of emergency hospital admission should be monitored routinely.

What's new?

  • Measures of the incidence of diabetes complications are needed to improve the quality of diabetes care, and emergency hospital admission data are being used for this purpose by national health agencies.
  • Modifiable risk and healthcare factors associated with these complications indicate how they may be reduced.
  • In the population in England, better scheduled access to primary healthcare and better glycaemic control were associated with lower emergency hospital admission rates for diabetes among patients with acute and chronic complications, admissions without complications, and hypoglycaemia.
  • A global measure of adverse diabetes outcomes resulting in hospital admission should be developed.

Introduction

Worldwide diabetes prevalence is expected to increase from 2.8% in 2000 to 4.4% in 2030 [1]. In England, there were an estimated 3.1 million people (7.4%) aged ≥16 years with diabetes in 2010, and the number is projected to rise to 4.6 million (9.5%) by 2030 [2].

Diabetes complications are common and result in poorer quality of life and reduced survival. The incidence of long-term microvascular complications of diabetes has declined in recent decades [3], but trends in macrovascular complications of diabetes are less positive. In developed countries, there has been a decline in mortality from cardiovascular disease in the general population over three decades. Although some studies show similar or greater declines in adults with diabetes [4, 5], cardiovascular disease mortality rates remain much higher. There have been relatively few recent studies that have included other endpoints, but in England between 2004 and 2008, the number of lower limb amputations among people with diabetes increased, whereas it decreased in the population without diabetes [6]. There is also evidence that severe hypoglycaemia in people with diabetes is associated with adverse outcomes [7]. In addition, acute complications of diabetes result in a large number of emergency hospital admissions [8].

Reducing complications and emergency hospital admissions for diabetes are therefore priorities in many countries. Associations between poor glycaemic control and adverse outcomes have been well documented [9]. Primary care-sensitive conditions, or, in the USA, ambulatory care-sensitive conditions, are diseases for which it is hypothesized that the risk of emergency hospital admission can be reduced by high quality primary care, including risk factor control. Diabetes is included in the primary care-sensitive condition list used by the National Health Service (NHS) Outcomes Framework [10], the Prevention Quality Indicators list of the US Agency for Healthcare Research and Quality [11], and the Organization for Economic Co-operation and Development Healthcare Quality Indicators [12].

In England, emergency hospital admissions for diabetes are generally to NHS-funded hospitals, and nearly all the population is registered with general practices. We investigated, at general practice level, the associations between hospital admission rates in four diabetes diagnostic categories (acute complications, chronic complications, no complications and hypoglycaemia), and primary healthcare resourcing (access, practice size and general practitioner supply) and hospital admission rates in four diabetes diagnostic categories (acute complications, chronic complications, no complications and hypoglycaemia), the Quality and Outcomes Framework (QOF); the general practice pay-for- performance programme), diabetes quality indicators and population factors (deprivation and ethnicity).

Methods

The present study did not require ethics approval as it was a secondary analysis of routinely available national data.

Data sources

Hospital Episode Statistics data

All NHS-funded hospitals in England provide hospital admission data to the Hospital Episode Statistics database, with diagnoses coded using the WHO's International Classification of Diseases 10th revision [ICD-10 (Table 1)]. In ICD-10, diabetes type is represented by the second and third digits, and there is no classification of diabetes control. We used primary care-sensitive condition definitions as a basis for the choice of complication groups that we included, except for hypoglycaemia, another diagnosis group which could be added to these definitions. We included in our analysis both Type 1 and Type 2 diabetes and all age groups. We used diabetes complications as a principal diagnosis only, in accordance with all primary care-sensitive condition definitions, rather than the overall burden of diabetes complications (Table 1). Many codes for chronic diabetes complications, in particular, may appear in subsequent diagnosis fields, with a code for the organ system affected as the primary diagnosis, but they are not hospital admissions for diabetes per se. Case-mix funding of NHS emergency admissions, implemented in 2006–2007, provided a financial incentive to hospitals to increase the coding of diabetes complications as secondary diagnoses, as they may receive additional payment for them. Hospital admissions with a major complication of diabetes, such as myocardial infarction or amputation, are likely to be coded with the complication rather than diabetes as the principal diagnosis. Hypoglycaemia is currently not part of primary care-sensitive condition definitions, but we think it should be, as it is an acute complication of diabetes treatment, so we included it in the present analysis.

Table 1. International Classification of Diseases 10th revision (ICD-10) diabetes diagnostic categories
  1. a

    Categories of complications we used.

ICD-10 three character codesE10 Insulin-dependent diabetes mellitus (Type 1)
E11 Non-insulin-dependent diabetes mellitus (Type 2)
E12 Malnutrition-related diabetes mellitus
E13 Other specified diabetes mellitus
E14 Unspecified diabetes mellitus
E16 Hypoglycaemiaa
E87 Acidosis (this category includes lactic acidosis as an acute problem in diabetes, but also acidosis not otherwise specified, metabolic acidosis and respiratory acidosis, so is not useful for diabetes hospital admissions)

ICD-10 4th character/extension

(The fourth-character subdivisions are for use with categories E10–E14 only)

Short-term complicationsa.0Coma (hyperosmolar/hypoglycaemic/hyperglycaemic)
.1Ketoacidosis
Long-term complicationsa.2 +  Renal complications (diabetic nephropathy, intracapillary glomerulonephrosis, Kimmelstiel–Wilson syndrome)
.3 +  Ophthalmic complications (cataract, retinopathy)
.4 +  Neurological complications (amyotrophy, mononeuropathy, autonomic neuropathy, polyneuropathy)
.5Peripheral circulatory complications (gangrene, ulcer, peripheral angiopathy)
.6Other specified complications (arthropathy, neuropathic)
.7Multiple complications
.8Unspecified complications
No complicationsa.9No complications

We calculated the indirectly age-/sex-standardized hospital admission rates for each practice using practice codes in the Hospital Episode Statistics data and practice population data provided by the NHS Information Centre. We did not distinguish between initial admissions or re-admissions as primary care-sensitive condition definitions do not make this distinction.

Quality and Outcomes Framework data

The Quality and Outcomes Framework (QOF), the pay for performance programme for general practices, QOF uses data from practice information systems to reward a range of process and outcome indicators of chronic disease care. We used practice-level QOF data for the years 2004–2005 to 2009–2010 including data on GP-registered prevalence, as this may affect admission rates [13]. The diabetes clinical quality domain includes eight indicators, of which the five ‘intermediate outcome’ indicators used were: Diabetes 6: the percentage of people with diabetes whose last HbA1c measurement was ≤57 mmol/mol (7.4%); Diabetes 20: the percentage of people with diabetes whose last HbA1c measurement was ≤59 mmol/mol (7.5%); Diabetes 23: the percentage of people with diabetes whose last HbA1c measurement was ≤53 mmol/mol (7.0%); Diabetes 12: the percentage of people with diabetes whose last measured blood pressure was ≤145/85 mm Hg; and Diabetes 17: the percentage of people with diabetes whose last measured total cholesterol was ≤ 5 mmol/L.

We defined the HbA1c variable as at or below target threshold. Because the definition of the QOF indicator target for glycaemic control has changed very slightly over time from ≤7.5% to 7.0% and back to ≤ 7.5%, we assumed identity between the HbA1c indicators Diabetes 6 (the first HbA1c target), Diabetes 20 and Diabetes 23 (the current target) to obtain a complete time series. As QOF data are aggregate we were unable to harmonize these thresholds.

To measure broad access to care from patients’ perspective, we used two indicators from the QOF Patient Experience domain: Patient Experience 07, patient experience of access (1): percentage of people who, in the national survey, indicated that they were able to obtain a consultation with their general practitioner within 2 working days and Patient Experience 08, patient experience of access (2): percentage of people who, in the national survey, indicated that they were able to book an appointment with their general practitioner > 2 days ahead.

Practice size and resourcing

The NHS Information Centre provided data on general practitioners per 100 000 registered patients, for the years 2004–2005 to 2009–2010, and an annual age/sex breakdown of practice populations, which was used to indirectly standardize hospital admission rates (expected counts). Practices with a list size of < 500 people (99/8300) were excluded as they were likely to serve atypical populations.

Population data

We used a time series of English Index of Multiple Deprivation weightings for each general practice for 2004, 2007 and 2010, produced by aggregating the postcodes of individual registered people [14]. Diabetes prevalence in a practice is likely to affect admission rates, so QOF prevalence was included as a covariate. As ethnicity is an independent risk factor for adverse outcomes in ethnic minority populations (controlling for prevalence) [15], we produced estimates of the proportion of South-Asian and black ethnic minority populations for each practice from ethnicity breakdowns of Hospital Episode Statistics data. This method has been externally validated [16].

Statistical analysis

We used negative binomial distribution regression rather than Poisson because there was over-dispersion of the data. From this, we obtained incidence rate ratios, which in this context are hospital admission rate ratios. Because of lack of independence of the data, we adjusted for the clustering effect of general practice. We assessed correlation coefficients, fitted bivariate and then multivariate models, and selected covariates using reverse stepwise selection. Because we used robust standard errors, which are forced by the use of cluster options, we have shown Wald-test results to evaluate model goodness-of-fit. Stata 11 was used for analysis.

Results

Characteristics of practice populations and percentage achievement of QOF indicators are shown in Table 2. After removing small (<500 registered population) practices, the study covered 54 741 278 people registered with 8140 (94% of 8622) general practices. For the QOF indicators chosen, scores were generally not as high as for some other care process indicators; for example, in 2009–2010 the mean practice score for QOF Diabetes indicators 6, 20 and 23, was only 53.82%, having fallen by almost six percentage points over 5 years. By contrast, the mean achievement of blood pressure and lipid control increased by ~10% over the study period. The mean prevalence of registered diabetes in England increased from 3.34% in 2004–2005 to 5.46% in 2009–2010, although some of this increase is attributable to the numerator and denominator changing from all people with diabetes and the whole population to only those aged > 17 years in both cases in 2007–2008, because prevalence increases with age. Over the study period, the mean total diabetes complication hospital admission rates (covering all four categories) fell from 73.30 to 68.51 per 100 000.

Table 2. Admission rates and aggregate characteristics of each practice population in 2004 and 2009
 20042009% change P a
MeanInterquartile range MeanInterquartile range
  1. a

    Mann–Whitney test for differences between means.

  2. b

    Hospital admission rates are adjusted for age and sex (indirectly standardized), calculated as [standardized hospital admission ratio (observed hospital admission rate indicators counts/expected hospital admission counts)] x [national hospital admission rate/100,000]. Indirect standardization was used because admission numbers are small at general practice level.

  3. Weighting for each practice was produced by aggregating Index of Multiple Deprivation scores from postcodes of individual registered patient.

  4. QOF Patient Experience indicator 07, percentage of people who, in the national survey, indicated that they were able to obtain a consultation with their general practitioner within 2 working days.

  5. QOF Patient Experience indicator 08, percentage of people who, in the national survey, indicated that they were able to book an appointment with their general practitioner > 2 days ahead.

  6. QOF Diabetes indicator 12, percentage of people with diabetes whose last blood pressure measurement was ≤145/85 mmHg.

  7. QOF Diabetes indicator 17, percentage of people with diabetes whose last measured total cholesterol within the previous 15 months was ≤5 mmol/l.

  8. QOF Diabetes indicators 6, 20 and 23, percentage of people with diabetes whose last HbA1c measurement was ≤53 mmol/mol (7.0%; or equivalent test/reference range depending on local laboratory) in the previous 15 months.

Observed hospital admissions/100 000 population
Total hospital admissions with diabetes73.3033.69–99.7268.5111.45–101.39−6.51%<0.001
Acute complications20.760.00–30.8719.850–29.42−4.38%<0.001
Chronic complications12.770.00–19.5810.120–15.02−20.75%<0.001
No complications25.200.00–38.2521.870–34.48−13.21%<0.001
Hypoglycaemia14.570.00–23.0716.660–26.5314.34%0.0018
Age-/sex-standardized hospital admissions/100 000 population a
Total hospital admissions74.3933.79–100.6969.4511.69–102.88−6.64%<0.001
Acute complications20.800–31.0919.830–29.52−4.66%<0.001
Chronic complications13.320–19.9010.540–14.95−20.87%<0.001
No complications25.430–38.1822.040–34.87−13.33%<0.001
Hypoglycaemia14.950–23.0717.510–27.2217.12%<0.001
Covariates
Deprivationb23.4313.18-31–7923.7713.66–32.051.45%0.019
Asian population (%)6.120.33–4.947.970.53–6.9530.23%<0.001
Black population4.000.21–3.484.670.28–4.7016.75%<0.001
Diabetes prevalence (%)3.432.83–3.915.604.66–6.3563.27%<0.001
General practitioner supply/100 000 population52.7843.82–59.6756.4745.39–64.966.99%<0.001
Practice list size6308.393191–86136724.973437.50–9192.506.60%<0.001
QOF Patient Experience indicator 07 (%)84.3279.5–93.0282.0776.27–91.15−2.67%<0.001
QOF Patient Experience indicator 08 (%)77.1567.47–90.5274.7764.89–87.60−3.08%<0.001
QOF Diabetes indicator 12 (%)70.1763.50–78.1380.8976.19–86.2715.28%<0.001
QOF Diabetes indicator 17 (%)71.3764.85–79.3182.8079.20–86.9116.02%<0.001
QOF Diabetes indicators 6, 20 and 23 (%)58.7351.70–65.9153.8248.48–59.32−8.36%<0.001

Regression modelling

Table 3 shows the results of the negative binomial multivariable regression analysis for hospital admissions for all the complication groups. The covariates for general practioner supply and achievement of total cholesterol targets were dropped from the model so those results are not shown. The effect sizes, as measured by incidence rate ratios, were generally small, apart from year and diabetes prevalence. Because of this, we have reported results to four decimal places.

Table 3. Stepwise multivariable negative binomial regression analysis for total hospital admissions with diabetes as the primary diagnosis
 Incidence rate ratioa P 95% CI
  1. Probability > chi-square < 0.0001.

  2. n/a  =  dropped from model after stepwise variable selection process.

  3. QOF Patient Experience indicator 07, percentage of people who, in the national survey, indicated that they were able to obtain a consultation with their general practitioner within 2 working days

  4. QOF Patient Experience indicator 08, percentage of people who, in the national survey, indicated that they were able to book an appointment with their general practitioner > 2 days ahead

  5. QOF Diabetes indicator 12, percentage of people with diabetes whose last blood pressure measurement was ≤145/85 mmHg.

  6. QOF Diabetes indicator 17, percentage of people with diabetes whose last measured total cholesterol within the previous 15 months was ≤ 5 mmol/l.

  7. QOF Diabetes indicators 6/20/23, percentage of people with diabetes whose last HbA1c measurement was ≤ 53 mmol/mol (7.0%; or equivalent test/reference range, depending on local laboratory) in the previous 15 months.

  8. a

    IRR is incidence rate ratio.

Deprivation1.0154<0.0011.0141–1.0166
South-Asian population0.9919<0.0010.9902–0.9935
Black population0.9937<0.0010.9918–0.9955
Diabetes prevalence1.0956<0.0011.0677–1.1241
General practitioner supplyn/an/an/a
Practice list size0.99990.0130.9999–0.9999
QOF Patient Experience indicator 070.99890.0230.9979–0.9998
QOF Patient Experience indicator 080.99880.0030.9980–0.9996
QOF Diabetes indicator 121.0031<0.0011.0017–1.0046
QOF Diabetes indicator 170.99870.0920.9971–1.0002
QOF Diabetes indicators 6/20/230.9971<0.0010.9958–0.9984
2005 (vs 2004)0.99700.7970.9741–1.0204
2006 (vs 2004)0.9102<0.0010.8735–0.9484
2007 (vs 2004)0.93130.0030.8879–0.9768
2008 (vs 2004)0.9061<0.0010.8585–0.9563
2009 (vs 2004)0.8393<0.0010.7856–0.8966

Increasing deprivation score, diabetes prevalence and achievement of target blood pressure are risk factors for hospital admission, while most healthcare covariates are protective; larger practice population, better access to care measures and better HbA1c target achievement showed significant associations with lower hospital admission rates. Higher proportions of South-Asian and black population are protective. There is a change over time, with the incidence rate ratio compared with 2004–2005 falling steadily over time since the QOF programme commenced.

Table 4 shows the multivariable practice-level analysis for complication categories. Results are generally consistent with those for total hospital admissions for some variables; for example, deprivation and diabetes prevalence remain risk factors for hospital admission across all categories. Higher percentage achievement of the HbA1c target and the ability to book an appointment are protective in most models. In contrast, higher percentage achievement of the blood pressure target is a risk factor for hospital admission in all models. There is no evidence that achievement of total cholesterol ≤5 mmol/l is associated with hospital admission rates.

Table 4. Stepwise multivariable negative binomial regression analysis for categories of hospital admissions with diabetes as the primary diagnosis
CovariatesHospital admissions with acute complicationsHospital admissions with chronic complicationsHospital admissions without complicationsHospital admissions with hypoglycaemia
 Incidence rate ratioa P 95% CIIncidence rate ratio P 95% CIIncidence rate ratio P 95% CIIncidence rate ratio P 95% CI
  1. n/a, dropped from model after stepwise variable selection process.

  2. Ind. Patient Experience 07, percentage of people who, in the national survey, indicated that they were able to obtain a consultation with their general practitioner within 2 working days.

  3. Ind. Patient Experience 08, percentage of people who, in the national survey, indicated that they were able to book an appointment with their general practitioner > 2 days ahead.

  4. Ind. Diabetes 12, percentage of people with diabetes in whom the last blood pressure was ≤145/85 mmHg.

  5. Ind. Diabetes 17, percentage of people with diabetes whose last measured total cholesterol within the previous 15 months was ≤5 mmol/l.

  6. Ind. Diabetes 6,20 and 23, percentage of people with diabetes whose last HbA1c measurement was ≤53 mmol/mol (7.0%; or equivalent test/reference range depending on local laboratory) in the previous 15 months.

  7. a

    IRR is incidence rate ratio.

Deprivation1.0160<0.0011.0140–1.01791.0119<0.0011.0096–1.01411.0138<0.0011.0122–1.01541.0226<0.0011.0206–1.0245
South-Asian population0.9842<0.0010.9818–0.98670.99530.0010.9925–0.99810.9947<0.0010.9923–0.9971n/an/an/a
Black population0.9908<0.0010.9879–0.99371.00380.0081.0010–1.00660.9897<0.0010.9872–0.99210.99580.0040.9929–0.9987
Diabetes prevalence1.1085<0.0011.0715–1.14671.05380.0221.0074–1.10221.0865<0.0011.0415–1.13331.05030.0051.0151–1.0867
GP supply n/an/an/an/an/an/an/an/an/an/an/an/a
Practice list sizen/an/an/a0.99990.0070.9999–0.99990.99990.0010.9999–0.9999n/an/an/a
Ind. Patient Experience 070.99800.0040.9967–0.9994n/an/an/an/an/an/a0.99790.0010.9967–0.9991
Ind. Patient Experience 08n/an/an/a0.9973<0.0010.9962-0.99840.9985<0.0010.9977–0.9993n/an/an/a
Ind. Diabetes 121.00310.0141.0006–1.00551.00290.0161.0006-1.00531.00260.0021.0009–1.00421.00330.0011.0014–1.0053
Ind. Diabetes 170.99750.0670.9949–1.0002n/an/an/an/an/an/an/an/an/a
Ind. Diabetes 6, 20 and 230.99660.0030.9943–0.99880.9910<0.0010.9888–0.99330.9970<0.0010.9954–0.99851.00250.0061.0007–1.0042
2005 (vs 2004)1.01780.4090.9760–1.06151.01930.4380.9713–1.06960.98070.2400.9495–1.01300.97450.2340.9339–1.0169
2006 (vs 2004)0.92590.0190.8682–0.98741.00880.8190.9361–1.08710.90280.0010.8484–0.96060.90050.0010.8453–0.9594
2007 (vs 2004)0.94400.1160.8784–1.01441.03280.4610.9479–1.12530.92400.0370.8578–0.99530.93800.0840.8724–1.0087
2008 (vs 2004)0.92390.0520.8530–1.00081.00960.8470.9161–1.11280.89160.0080.8194–0.97020.92370.0510.8529–1.0003
2009 (vs 2004)0.8355<0.0010.7587–0.92000.87990.0310.7834–0.98820.83820.0010.7555–0.92990.92820.1050.8482–1.0158
 Probability chi-square < 0.0001Probability chi-square < 0.0001Probability chi-square < 0.0001Probability chi-square < 0.0001

Discussion

Main findings

In the present observational study of factors associated with hospital admissions for diabetes complications as a primary diagnosis, the population factors practice deprivation and registered diabetes prevalence were associated with higher diabetes admission rates for all categories of diabetes complications. Hospital admissions reduced with time overall, but only for 2009–2010 compared with 2004–2005 for most diagnostic categories. The effect sizes were in most cases small. For example, increasing Diabetes 6, 20 and 23, the percentage of people with diabetes whose last HbA1c measurement was ≤53 mmol/mol (7.0%), by 1% would reduce the admission rate by only 0.29%.

Strengths and limitations

A strength of the present study was its 5-year period of follow-up after QOF implementation, and the fact that we were able to analyse categories of diabetes complications. Annual admission counts per practice were relatively low, giving low median rates for acute, chronic and hypoglycaemia hospital admission categories. We used appropriate statistical methods to deal with this.

A limitation of the present analysis is that it would have been desirable to undertake it at an individual patient as well as at a practice level. Demonstrating an association between diabetes control and admission rates at the practice level does not necessarily mean that the same association exists at patient level. Because the QOF is primarily a payment system, it is practice- not patient-based, and while the UK has several large patient-level general practice research databases, individual practices are not identified in them. This prevents linkage of patient data with healthcare quality variables such as QOF indicator achievement and practice resourcing. Accuracy of diagnostic coding of diabetes is an important limitation, but to be consistent with primary care-sensitive condition definitions we only considered principal diagnoses, and 87.2% of these are coded correctly [17]. Finally, practice nurses provide a significant amount of diabetes care, but data on practice staffing is not available at national level.

Total cholesterol was used instead of LDL cholesterol, which is a primary lipid target in diabetes care, because the latter is not available through the QOF. Because these physiological or intermediate outcome indicators are harder to achieve than care processes because they involve patient factors, there is less of a ‘ceiling effect’ (clustering of practice achievement values around a high achievement percentage) than with clinical process indicators.

We included the proportion of the practice population which was black and South Asian in addition to prevalence because healthcare inequalities in these ethnic groups have been documented previously. Both these covariates are protective in most hospital admission categories. These findings are consistent with evidence that diabetes care in South-Asian people has improved in recent years, and that care processes are now similar to those in the rest of the population, although outcomes may still be poorer [15, 18]. Among black people this finding may also be attributable to increased registration of less severe undiagnosed cases in recent years, in a population that previously did not access preventive healthcare as effectively.

The present study examined primary diagnoses only. While this is a very important element of diabetes complications and is the focus of international and national primary care-sensitive condition indicators, developing an overall diabetes complications categorization that also included primary diagnoses to which diabetes was a major contributor would be a more useful global outcome measure. This would also overcome the limitation of a possible coding shift over time from diabetes as a primary diagnosis to a secondary diagnosis. Current primary care-sensitive condition definitions do not take sufficient account of the influence that primary care could plausibly have.

Comparison with other studies

We found that people's perception of ability to book a non-urgent appointment with their general practitioner > 2 days ahead, which is important for ongoing chronic disease management, was associated with a reduced hospital admission risk. In contrast, a US study found only a modest effect of long wait times on primary care utilization, and no robust effect on health outcomes, including hospital admission [19]; however, they did not control for other aspects of care quality, and we have documented similar associations for other diseases [20]. The present analysis suggests that patient perceptions of access may be clinically important, as they may influence people to seek emergency care through Emergency Departments rather than from their general practice. A larger practice list size may be a marker for better organization of care and availability of staff.

We found that the percentage of people whose last HbA1c measurement was ≤ 53 mmol/mol (7.0%) was protective for all diagnostic categories except hypoglycaemia, but achievement of this goal deteriorated from 58 to 53% between 2004 and 2009. The HbA1c threshold was lowered in 2009–2010, which will have lowered achievement slightly. Hyperglycaemia generally increases over time, but the QOF has no patient-level longitudinal data or data on years since diagnosis. Another report found some associations at practice level between admission rates for ‘short-term complications’ and three QOF indicators of glycaemic control [21]. This analysis covered 2001–2002 to 2006–2007 (2004–2005 was the first year of the QOF), while we analysed time trends to 2009–2010 to better measure its effects.

The influential Action to Control Cardiovascular Risk in Diabetes (ACCORD) study, published in 2008, showed that the use of intensive therapy to target normal HbA1c levels for 3.5 years increased mortality and did not significantly reduce major cardiovascular disease events [22], and that symptomatic, severe hypoglycaemia was associated with an increased risk of death [23]. These findings may have led to less subsequent clinical effort to reduce HbA1c to near-normal levels. A new meta-analysis is consistent with earlier evidence that the cardiovascular disease benefit of intensive glucose lowering seems to be modest, and that glucose lowering is probably less efficacious and more difficult to achieve than lipid and blood pressure control [24].

Another meta-analysis showed that statins are effective at decreasing adverse outcomes [25], so a combined approach that targets glucose lowering, lipid lowering, and blood pressure control has been proposed [26]. We found no evidence that a high percentage of patients with total cholesterol ≤5 mmol/l was associated with admission rates; however, cholesterol control is more likely to be associated with admissions for cardiovascular disease complications rather than for diabetes. Although most people with diabetes are treated with statins, diabetes is not a disease associated with high total cholesterol, so many people have values < 5 mmol/l without specific intervention.

A higher percentage of people with diabetes whose last blood pressure was ≤145/85 mmHg is a risk factor for hospital admission in all categories. Recent studies have found that that targeting near-normal levels of blood pressure has limited benefit, with most benefit being achieved by targeting a level of <140 mmHg [27]. A meta-analysis showed that, with more aggressive goals (<130 mm Hg), the risk of stroke continued to fall, but there was no reduction in risk of other events, and the risk of other serious adverse events increased [28]. A lower blood pressure in older people with diabetes may also be associated with increasing cardiovascular disease morbidity and mortality risk [29]. Finally, like cholesterol, blood pressure control may mainly affect other reasons for emergency hospital admission such as stroke.

Time since 2004–2005 was generally strongly associated with lower admission rates, especially in the overall analysis. As our time series does not extend before QOF started, we cannot be sure that this effect was because of QOF or other unmeasured covariates. A previous analysis over the period 2000–2007 reported that QOF had no discernible effects on hypertension-related processes of care or clinical outcomes, which improved over time [30]. Registered diabetes prevalence increased during the observation period. This may have resulted in many uncomplicated new cases being identified. Once on registers, these people will have diluted complication rates.

Conclusions

Most effect sizes for clinical indicators were not large, but might be greater in countries where primary healthcare is less well developed. Conversely, our analysis suggests that the potential for significant reductions in diabetes emergency hospital admissions by further improving the clinical quality of primary care may be limited in countries where it is well-developed. Other initiatives such as patient education or integrated care may have a greater impact. For example, the incidence rate ratio for the proportion of people whose HbA1c is at target is 0.9971, i.e. an increase of 1% of those reaching this target is associated with a 0.29% decrease in hospital admissions for diabetes. In 2009, there were 410 940 hospital admissions with a primary diagnosis of diabetes. A 0.29% decrease represents only 1192 hospital admissions across England.

We found primary care factors had an effect on all four of our categories of complications, suggesting that there is little rationale in restricting primary care-sensitive condition definitions to acute complications. Only the US Agency for Healthcare Research and Quality indicators include a chronic complications category [11]. Primary care-sensitive condition definitions for diabetes should be reviewed to take account of our findings, and should include admissions for hypoglycaemia and ‘no complications’ categories, as they are as numerous as acute complications admissions.

In conclusion, there are many difficulties involved in using routine data sources to monitor diabetes epidemiology and outcomes, but as they are being widely used for these purposes, improved coding, accurate and harmonized definitions, and careful analysis are essential. From a practical, clinical perspective, it is important firstly to monitor emergency hospital admission rates as an outcome indicator of ongoing clinical care, and secondly to use the broader categories we have described here, rather than just acute complications of Type 1 and Type 2 diabetes.

Funding sources

The study was funded by the NHS Institute for Improvement and Innovation, which agreed the overall study design, but had no other role.

Competing interests

None declared.

Acknowledgements

We thank the Association of Public Health Observatories and the NHS Information Centre for Health and Social Care for supplying some of the data used in this study.

Ancillary